Harness Hub

The directory · 286 startups

Startups

Teams building the harness layer, from orchestration frameworks and memory systems to sandboxes, evals and vertical agent products. Each entry notes where they are based and when they started.

All startups

Featured

Devin, the AI software engineer, and the Windsurf IDE.

Cognition builds Devin, an autonomous AI software engineer that plans and executes engineering tasks end-to-end, and acquired the Windsurf agentic IDE in 2025. One of the defining startups of the coding-agent wave.

~$10.2B valuation (Sep 2025)

3 pros · 4 cons
Featured

The de-facto standard toolkit for building LLM applications and agents.

LangChain makes the LangChain framework and LangGraph, the low-level orchestration standard for stateful, controllable agents, plus the LangSmith platform for tracing, evaluating and monitoring them in production.

~$75M raised

4 pros · 4 cons
Featured

The open platform for cloud coding agents.

OpenHands (formerly All Hands AI, born from the OpenDevin project) builds the major open-source coding-agent platform — agent canvas, sandboxed execution, SDK and an enterprise control plane — now rebranded under the project's name.

4 pros · 4 cons
Featured

Conversational AI agents for customer experience.

Sierra, co-founded by Bret Taylor and Clay Bavor, builds production conversational agents that resolve real customer issues with guardrails, evaluation and human escalation — a template for deploying vertical agents at the enterprise edge.

~$10B valuation (Oct 2025)

4 pros · 4 cons

An orchestration layer for coding agents such as Claude Code and Codex, from YC-backed 21st.

1code is an open-source orchestration layer for coding agents, built by 21st, a Y Combinator W26 company that YC lists as making "UI building blocks for the agentic internet". The 1code repository describes itself as an orchestration layer for Claude Code and Codex, so several agents can be run and supervised together.

A "Software Factory" for enterprise teams that wants production-quality software, not vibe-coded prototypes.

8090 builds an AI-native software development platform it calls the Software Factory, aimed at enterprise teams that need "production-quality software, not just vibe-coded prototypes" and taking work from spec to deploy. Founded in January 2024, it raised a $135M Series A led by Salesforce Ventures in June 2026, according to TechCrunch.

$135M Series A led by Salesforce Ventures (Jun 2026)

3 pros · 3 cons
Acquired

ACT-1 and the original computer-use agents — team joined Amazon.

Adept pioneered general-purpose agents that operate software via ACT-1/ACT-2. In 2024 much of the team joined Amazon's AGI lab — an early, emblematic consolidation of the harness layer into big tech.

3 pros · 4 cons

Identity Control Plane for AI Agents

Aembit provides runtime IAM for AI agents and workloads. It enforces access policies, keeps downstream credentials out of agents, audits access, and can revoke access with a kill switch.

Isolated cloud sandboxes that host Hermes, OpenClaw, Claude Code, Codex, OpenCode, Pi and QM.

Agent37 hosts many agent harnesses — Hermes, OpenClaw, Claude Code, Codex, OpenCode, Pi and YC's QM — in isolated cloud sandboxes billed by the minute; its FAQ prices a 2 vCPU / 4 GB sandbox running all month at $4.76. Its founder says it "started as one of those OpenClaw wrapper companies back in February". Launch HN in October 2026 drew 163 points; Y Combinator F26.

An infinite-canvas desktop workspace for running parallel coding agents.

AgentGrid is a desktop app for macOS, Windows and Linux that puts multiple coding agents (Claude Code, Codex, Cursor, OpenCode and others) on one infinite canvas alongside terminals, browsers and notes. A Coordinator pane takes natural-language instructions to create agents and teams, such as a builder and a QA reviewer, arrange the workspace and direct work across the canvas. Users bring their own AI subscriptions, and the canvas is shared so teams can follow every agent and review results together.

Kanban board that hands tasks to your agent harness with persistent context

AgentKanban is a kanban board for planning work and handing tasks off to AI agents, keeping agent chat activity attached to each task. Claude Code, Codex, Cursor or any MCP-compatible agent can load a task by ID, resume it with context and update the board. It also ships a VS Code extension.

Email Inbox API for AI Agents

AgentMail is email infrastructure built for AI agents. Agents get their own inboxes and can send, receive and act on email at scale.

Agent observability, testing and session replay.

AgentOps instruments agents to record every step, cost and tool call — then lets teams replay sessions and build regression tests from real failures.

3 pros · 4 cons

A self-hosted control layer for fleets of CLI coding agents.

AgentsMesh calls itself an 'AI Agent Workforce Platform' for running many coding agents across your own machines from one console. Self-hosted runners report their capacity and schedule agent pods, each with its own Git worktree sandbox, credentials and branch. Web, Electron desktop and iOS clients stream terminals in real time, and an Autopilot control agent sends each idle pod its next instruction, with iteration caps, decision history and human takeover. Supported agents include Claude Code, Codex CLI, Gemini CLI, Aider and OpenCode. The Go codebase is source-available under BSL-1.1, converting to GPL-2.0-or-later.

The programming language for agentic software.

Agno (formerly Phidata) builds a high-performance open-source agent runtime — memory, knowledge, storage and monitoring built in — aiming to be the substrate beneath agentic applications rather than just a framework.

3 pros · 3 cons

Monitoring and continuous improvement for AI agents: detects failures and opens fix PRs.

Agnost AI launched on Hacker News in July 2026 with product analytics drawn from agent conversations, such as rage-prompting and rephrasing. Its site now reads "Monitoring and Continuous Improvement for AI Agents": it detects failures and opens pull requests that fix them. Y Combinator S26.

AIR

An inline firewall for AI agents and the skills, plugins and MCP servers they use.

AIR builds what it calls an inline firewall for AI agents and the skills, plugins and MCP servers they rely on, sitting between agents and their add-ons to stop malicious behaviour. It came out of stealth on 1 September 2026 with $50M led by Sequoia and Greenoaks.

$50M led by Sequoia and Greenoaks (Sep 2026)

4 pros · 4 cons

Discover, secure, govern, and optimize every AI agent.

Airia is an enterprise AI security and governance platform that discovers the models, agents, tools and MCP servers a company uses, including unapproved ones. It lets teams build and run agents across models and frameworks, and enforces agent actions at runtime by allowing, blocking, requiring approval, redacting context or logging before a tool call runs.

Privileged access management for AI agents, with per-end-user identity instead of one shared key.

Alter is a credential proxy that gives multi-user agents per-end-user identity and access control. Its founders describe the problem as agents that connect "with one shared API key, so the agent can do anything that the key can do." The site now presents Alter as privileged access management for AI agents. Y Combinator S25.

AWS platform for building, connecting and running production AI agents with any framework or model

Amazon Bedrock AgentCore is AWS's platform for building, connecting and running production AI agents with any framework or model. It includes a harness for deploying agents and a Gateway that handles authentication and access control for tool calls to MCP servers, APIs and Lambda functions. It also has observability and evaluation tools for tracing agent behavior and testing variations against real traffic.

AWS's agentic coding assistant across IDE, CLI and console.

Amazon Q Developer is AWS's generative-AI assistant for software development, available in IDEs, the command line, the AWS Console and chat apps such as Slack and Teams. Its agentic capabilities write and test code, review and refactor it, run upgrades such as Java version and .NET Windows-to-Linux porting, and can read and write files, generate diffs and run shell commands. AWS states it will discontinue support for the Amazon Q Developer IDE plugins on April 30, 2027, pointing users to Kiro for similar capabilities.

Multiplayer cloud workstations for coding agents and humans.

Amika's 'Rigs' are cloud VMs loaded with your dev setup and favorite coding agents — booted in seconds, steered from web, Slack, Linear, SSH, CLI or API, and shareable so teammates can inspect previews or take over an agent session. Open core.

Amp

Frontier coding agent that runs each thread in its own remote machine.

Amp is a coding agent and development environment that spun out of Sourcegraph in December 2025 to become an independent company, with an announcement signed by its co-founders including Quinn Slack and Beyang Liu. Each Amp agent thread runs in a remote machine called an orb with your code, plugins and tools, so work continues unattended; orbs can wake on Linear issues, failing CI or schedules, be shared with teammates, and, since October 2026, be self-hosted. Amp also offers a CLI (including a headless mode for servers and CI), subagents, code review, Slack and Linear integrations, and an MCP-exposed feature.

Cloud-hosted browser infrastructure for computer-use and browser agents

Anchor describes itself as a cloud-hosted browser that lets AI agents use the web the way a person would. Features include Anchor Chromium (a stealth Chromium fork), OmniConnect for login handling, a built-in VPN and a Web Action Cache that turns planned tasks into deterministic code. It can be deployed in the customer's own cloud and is reachable through an SDK or MCP.

The company behind opencode, the open-source coding agent.

opencode — the terminal-first, model-agnostic coding agent with LSP integration, parallel sessions and share links — grew past 200k stars and moved from its original SST home to Anomaly (GitHub: anomalyco), which now steers the project behind opencode.ai.

3 pros · 4 cons
Acquired

Agent access governance — what agents can reach.

Apono built just-in-time access governance for cloud and data infrastructure, then extended it to AI agents — granting scoped, auditable permissions to agent identities instead of static credentials. Acquired by 1Password in July 2026 to anchor its agentic access-control line.

Acquired by 1Password (Jul 2026)

3 pros · 3 cons

Authorization and tool infrastructure for AI agents.

Arcade (arcade.dev) builds the identity and authorization layer for agents — OAuth-backed tool execution, credential management and fine-grained access control so agents can act on a user's behalf without inheriting god-mode credentials. Raised a $60M Series A in 2026.

$60M Series A (2026)

4 pros · 4 cons

Database sandboxes for you and your coding agents, so agents test against a realistic Postgres.

Ardent gives coding agents Postgres database sandboxes. Its founders argue that without "a realistic sandbox at the DB layer for testing, they ship garbage that can take down production databases." The Y Combinator Spring 2026 company launched on Hacker News in May 2026.

Sandboxes with stateful twins of the services agents act on

Arga Labs provides sandboxes for testing and training AI agents, with stateful twins of third-party APIs, CLIs and MCPs such as Slack, GitHub and Gmail. Teams point their agents at the twins, seed them with data and review full traces of every call and side effect.

Observability and evaluation for AI — makers of OSS Phoenix.

Arize traces and evaluates LLM and agent systems in production; their open-source Phoenix tracer is a common instrumentation choice for agent harnesses that need to be debuggable.

4 pros · 5 cons

AI governance for everyone

Arthur offers one control plane that brings together security, finance and the teams shipping AI. It measures, monitors and governs enterprise AI at scale.

Identity Security for AI Agents & NHIs

Astrix gives visibility and control over AI agents and non-human identities. It secures agent identities with governance, least-privilege access and full audit trails.

Always-on software delivery agents built on shared codebase context.

Augment Code (Augment Computing Inc.) builds AI coding agents for professional engineering teams. Its Cosmos platform is pitched as an always-on software delivery system from spec to verification, with pre-built agent loops for code review, ticket-to-PR, vulnerability remediation and incident response. Work is triggered from GitHub, GitLab, Slack, Linear, Jira, schedules or webhooks, runs in managed cloud sandboxes or customer-managed machines, and carries permissions, audit trails, observability and human review points, with per-task model choice. Founded in 2022 by Igor Ostrovsky and Guy Gur-Ari.

$252M total

3 pros · 4 cons

Ship AI agents faster and more securely

Auth0, part of Okta, offers identity tooling to help developers ship AI agents faster and more securely. The solution page is aimed at moving identity out of the developer's way while building agents.

A self-improving software factory with evaluator and healer meta-agents.

Autoheal runs always-on cloud agents for SDLC work such as incident response, vulnerability remediation, release readiness and AI coding cost control. Supervisory Evaluator agents score worker output using signals like review comments, CI failures and linked incidents, while Healer agents propose changes to prompts, skills, tools or model choice, test them against historical benchmarks and submit them for engineer approval in Git. Each run gets a fresh sandbox in the customer's cloud with policy checks on every tool call, per-agent budgets and audit trails. It raised a $7.9M seed led by Innovation Endeavors (Sept 2026).

$7.9M seed led by Innovation Endeavors (Sep 2026)

4 pros · 4 cons

Agentic Work Graph: task management for AI agents, with atomic claims, parallel lanes and proof of completion.

AWG is a work graph for coding agents: intent, requirements, tasks, dependencies, decisions, validation, evidence and outcomes in one shared source of truth across sessions, repositories, tools and machines. Agents claim work atomically (leases with fencing tokens), take parallel lanes that do not collide, and close a task only when its completion contract is met with evidence, not by assertion. People see what is ready, blocked, running, proven or waiting for a decision in one ledger. It runs locally on SQLite with no account, self-hosted on PostgreSQL, or as AWG Cloud at awg.palgu.ai with a Palgu account. A Claude Code mod and an MCP server for Codex and other clients expose tools such as awg_next, awg_submit and awg_create.

An agent-driven workbench for enterprise IT modernization and tech-debt reduction.

AWS Transform is an enterprise IT modernization workbench driven by AI agents for cloud migration, application modernization and ongoing tech-debt reduction. Its agents automate infrastructure migration from VMware, bare-metal and hybrid environments, and modernize mainframe, Windows and custom code. A continuous modernization capability, in preview, scans codebases for issues such as end-of-life dependencies and vulnerabilities, then prioritizes and fixes them.

Run many AI coding agents in parallel from one desktop app

Baton is a desktop app for running multiple AI coding agents at once, each in its own isolated git worktree, with progress monitoring, diff review and notifications. It supports Claude Code, Codex CLI, OpenCode, Antigravity CLI, Gemini CLI, Amp, Goose and other presets, plus custom terminal agents. It is free for up to 4 workspaces, and a paid license removes the limit.

Baz

AI code review platform for engineering teams.

Baz describes itself as an AI code review platform that helps engineering teams ship better code faster. It reviews pull requests through a GitHub app.

Sandboxes, persistent VMs and serverless GPUs for AI agents

Beam is a cloud platform for agents that offers persistent VMs, isolated sandboxes, inference endpoints and serverless GPUs. Sandboxes boot in seconds to run untrusted, agent-generated code, use custom images, and can fan out from one saved snapshot, for example for RL rollouts. The open-source runtime is beta9, and the site says the company is backed by Y Combinator.

Production infrastructure for AI agents

Bento is a production infrastructure platform for AI agents that monitors what runs, flags regressions and turns failures into fixes. It provides OpenTelemetry-native traces, natural-language alerts, drift detection, evaluations and versioning. The site says it is backed by Y Combinator.

p0: a desktop app that orchestrates coding agents to bring software development in-house.

BePurple, Inc. makes p0, a desktop app for macOS and Windows that turns a feature description into a spec and mockups and then orchestrates coding agents to build it. It runs on Claude Code or Codex, and on any endpoint compatible with the Anthropic Messages API, such as Kimi, DeepSeek, GLM or OpenRouter. It splits work into phases, runs agents in parallel with file ownership in isolated Git worktrees, and uses QA agents to check acceptance criteria before a PR opens. It also generates codebase Standards and has a Custom Agents Studio. Code generation runs locally, and model usage is billed by the user's provider.

The gateway between your harness and every model provider.

BerriAI builds LiteLLM — the open-source proxy that normalizes 100+ LLM providers behind the OpenAI API with routing, budgets and observability. A quiet standard inside production agent stacks.

3 pros · 4 cons

Codebase-aware model router for coding agents

Bito's Governor is a router that grounds each coding-agent request in your codebase and sends it to a right-sized model. It combines a code context engine with a complexity-scoring model router. It works with Claude Code, Cursor, Codex and GitHub Copilot through a base-URL change, and supports per-key budgets. Bito also offers a system-aware AI PR review agent.

Acquired

Perpetual-standby microVM sandboxes for production agents, now part of Baseten.

Blaxel (YC Spring 2025) provides compute, storage and networking for autonomous agents. Each sandbox is an isolated microVM that resumes in about 25ms with memory state intact and can stay on standby indefinitely when idle; it also offers batch jobs across sandboxes, MCP servers as first-class workloads, persistent volumes, egress controls and a model gateway. Founded in 2024 in San Francisco by Paul Sinai, Christophe Ploujoux and others, it was acquired by AI inference company Baseten in 2026 (terms undisclosed); YC lists it as acquired.

Autonomous enterprise software development with thousands of agents in parallel.

Blitzy builds an autonomous software development platform for large enterprises with sprawling legacy codebases. Per SiliconANGLE, it reverse-engineers existing environments into a dynamic knowledge graph of the codebase and uses an orchestration layer to run thousands of agents in parallel for days or weeks of inference, calling models from Google, Anthropic and OpenAI over 100,000 times in a single run. Founded in 2023 by Brian Elliott and Sid Pardeshi; it raised $200M at a $1.4B valuation led by Northzone in May 2026.

$200M at $1.4B valuation led by Northzone (May 2026)

3 pros · 4 cons

Build websites, apps and prototypes by chatting with AI

Bolt is a chat-based AI tool for building websites, apps and prototypes, with Bolt Cloud providing hosting, databases, authentication and integrations. Its Bolt Agent (Standard and Max tiers) automatically picks a model per task and tests, refactors and iterates to reduce errors.

The micro-VM for AI agents

BoxLite is a managed cloud that runs AI agents' apps in hardware-isolated micro-VMs, each with its own kernel and a public URL. It is built on an open-source runtime that can also run locally or on your own servers.

Enterprise-grade evals, data and AI gateway.

Braintrust turns prompts and agent traces into testable datasets and CI, plus an inference gateway — helping teams measure whether harness changes actually improve agent quality.

4 pros · 4 cons

The company behind the open-source web-agent library.

Browser Use maintains the eponymous 100k-star open-source library for vision-capable browser agents and commercializes it for enterprises automating web workflows.

4 pros · 4 cons

Headless browser infrastructure built for AI agents.

Browserbase runs fleets of headless browsers as an API — sessions, proxies, stealth, debugging — so agents can act on the web without operators managing browser infrastructure. Also builds the Stagehand SDK.

3 pros · 4 cons

Managed headless browsers for automation and AI agents, over MCP, Puppeteer or Playwright

Browserless runs headless browsers in the cloud that Puppeteer, Playwright and Selenium connect to over WebSocket. It offers an MCP server with scraping, crawling, search and export tools, BrowserQL (a GraphQL interface with stealth options), and REST endpoints for screenshots, PDFs and scraping. It lists integrations with Browser Use, Stagehand, LangChain and Claude Code.

A faster coding agent, from the Y Combinator Summer 2026 batch.

Bullet launched on Hacker News in August 2026 as "a faster coding agent", a new first-party coding harness competing on speed. The Launch HN post drew 121 points and 89 comments. The company is part of Y Combinator's Summer 2026 batch.

An IDE and dashboard for orchestrating fleets of cloud coding agents.

Capy says it was built from scratch for multi-agent orchestration. Developers assign feature, bug and refactor tasks to parallel coding agents, each on its own branch and sandboxed cloud VM, and the agents plan, build and open pull requests that are reviewed as normal. It ships desktop apps for macOS, Windows and Linux plus a web app, a Slack integration, automations, and end-to-end testing with native computer use, browser use and annotated videos. Capy (YC F24) was founded in 2024 in San Francisco by Nalin Semwal and Justin Sun.

Y Combinator (F24)

How AI Agents Use Money

Catena is a banking and governance platform for AI agents. It supports payments, treasury and other services.

Test, Monitor and Self Improve voice agents

Cekura is an automated QA platform for voice and chat AI agents. It runs synthetic conversations before an agent goes live, monitors production calls for drift, and can block deploys when release test suites fail. It also does adversarial red-teaming for jailbreaks and data leaks, and suggests prompt fixes for issues it reproduces in simulation.

Local validation for your agent's code

Chunk is a CircleCI tool that validates code written by Claude Code, Cursor or other agents before it is pushed to CI. It syncs local code to a remote Linux microVM sidecar and runs tests there as 'microbuilds' inside the agent loop, using hooks to give deterministic feedback. Sidecars can run integration suites with services like Postgres and Redis. It is free on every CircleCI plan.

Messaging between coding agents: every agent gets an address, across people, vendors and tools.

clawalkietalkie gives each coding agent an address, such as cc: for Claude Code, cx: for Codex, cu: for Cursor, gm: for Gemini CLI and gh: for GitHub Actions, so agents can hand work to each other, including a teammate's agent at a team address, and report what they are working on to one status board. CI can message the agent that can fix a failing build and wait for the reply. Claude Code's mod starts a turn when a message arrives; other agents read their inbox through an MCP server. The clients are open source (MIT) and talk to a hosted relay.

Security gateway for AI agents with purpose-based authorization

Clawvisor is a security gateway that sits between agents and services like email, calendar and code. It checks each request against the human-approved task, injects credentials server-side so agents never see them, and logs every action. It is source-available, self-hostable or managed, and backed by Y Combinator.

AI agent for production engineering

Cleric checks each change shipped to production and catches regressions that tests miss. It handles alerts, dismisses false positives and escalates with cause and evidence. For real issues it proposes a fix as a pull request and verifies the fix after deploy. It works in Slack or Microsoft Teams, with read-only access by default.

Open-source Ghostty-based macOS terminal for running many AI coding agents at once, by Manaflow.

cmux is a native macOS terminal, built in Swift and AppKit on libghostty, for multitasking across AI coding agents. It is made by Manaflow (YC Summer 2024), which its YC profile calls an open-source applied AI lab building next-gen devtools. Features include notification rings and a panel that flag panes needing agent attention, and vertical tabs showing git branch, PR status, ports and the latest notification. It also has a scriptable in-app browser, SSH workspaces, a one-command Claude Code teammate mode, and a CLI and socket API for automation. The terminal is GPL-3.0-or-later, and parts of the server code are BUSL-1.1.

Launched as an API for computer-use agents on legacy apps; now builds living worlds and multi-agent environments.

Coasty launched on Hacker News in July 2026 with an API for computer-use agents that operate legacy desktop and web apps. Its site now reads "Living Worlds & Multi-Agent Environments", a pivot towards environments where many agents act. Y Combinator S26.

The free coding agent, built on the open Codebuff multi-agent framework

Codebuff is a YC-backed terminal coding agent, installed with `npm install -g codebuff`. Its open repo now ships as Freebuff, built on Codebuff, which the README calls the open multi-agent framework behind its orchestration, tools and SDK (@codebuff/sdk). Specialized agents find files, implement, review, research docs and test in a browser, and can run in parallel in isolated workspaces.

Free iPhone companion for Claude Code and Codex sessions on your own Mac and Linux computers.

CodeConnect brings sessions from multiple paired Mac and Linux computers into one iPhone view. Both Claude Code and Codex support approval requests and messages. Claude Code also supports questions, diff review, hunk comments and a live terminal. Codex keeps its native approval choices and supports stopping a turn, without a phone terminal. Tailscale is the default connection; session content travels directly between the phone and computers. A separate push relay handles notification metadata, not session content. The daemon is MIT-licensed; iOS source is available under separate personal-build terms. The iPhone app is free and requires iOS 17 or later.

Agent that finds and ships performance optimizations as PRs

Codeflash's agent autonomously finds code optimizations for cost, latency, memory and throughput. Each change ships as a mergeable PR with benchmark numbers attached, after review by Codeflash's performance engineers. The company specializes in ML performance.

Acquired

AI coding agents that let non-engineering teams ship work, and an orchestration platform for them.

Codegen built AI coding agents that let non-technical teams carry out work that used to need engineering time, and an orchestration platform for running coding agents with task context. ClickUp acquired it in December 2025 to bring coding agents into ClickUp.

AI coding agent that thinks, plans, then builds

CodeGPT is an AI coding agent for VS Code and JetBrains IDEs. It reads the codebase, proposes an editable plan and writes code with step-by-step approval, can run terminal commands with approval, and connects to APIs, docs and databases via MCP. Users can bring their own keys for Claude, GPT, Gemini, Grok or local models via Ollama and LM Studio.

Your developers have guardrails. Your AI doesn't.

Coder provides self-hosted AI development infrastructure that lets developers use any model or agent while giving enterprises governance and control. Coder Agents is a native coding agent that works with any model on infrastructure you control. Coder AI Governance centralizes model access, enforces policy-controlled environments and keeps audit trails. Coder Workspaces are standardized, self-hosted environments shared by developers and agents.

AI code review and change governance for agentic development.

CodeRabbit runs automated AI review on every pull request, posting line-level findings and committable suggestions, and has expanded into what it calls agentic change management: PR triage that prioritizes the review queue against each reviewer's capacity, change explanations and continuous security monitoring of dependencies and code health. It positions itself as an independent review and control layer as AI-generated code outpaces human review capacity. In September 2025 it raised a $60M Series B led by Scale Venture Partners at a $550M valuation, with NVIDIA's NVentures participating.

$60M Series B led by Scale Venture Partners at $550M valuation (Sep 2025)

3 pros · 4 cons

The sessionless, relentless 24/7 personal agent.

Comma is an always-on personal agent with no session boundaries — 'no full stop until it's done'. Its distributed harness, Salix, loads tiny eBPF programs into the cluster to watch events around the clock and wake the main agent loop only when something needs it.

The coding agent for open models

Command Code is a terminal coding agent (`npm i -g command-code`) built for open-weight models such as DeepSeek, Kimi, GLM, Qwen and MiniMax. It also supports Anthropic, OpenAI, Google and xAI models. Its harness validates and repairs tool calls as they happen and targets high cache-hit rates. 'Taste' learns from accepts, rejects and edits and turns them into shareable project skills. It offers headless mode, a background sandbox, a desktop app, custom agents and MCP servers.

The integration layer: hundreds of managed tools your agents can call.

Composio handles auth, schemas, observability and reliability for the external SaaS tools an agent uses — 'plumbing for agentic actions' across Slack, GitHub, Salesforce and 250+ other surfaces.

4 pros · 4 cons

Run a team of coding agents in parallel isolated workspaces, locally or in the cloud.

Conductor creates isolated copies of a codebase, runs coding agents in each, and lets developers review, open PRs for and merge their work. It runs the first-party Claude Code, Codex, Cursor and OpenCode agents with the user's own subscriptions and keys. Conductor Cloud gives each agent a Firecracker microVM sandbox with repos and dependencies pre-installed, and workspaces are multiplayer. There are macOS and iOS apps and an API. Founded by Charlie Holtz and Jackson de Campos (YC S24), it raised a $22M Series A from Spark and Matrix in March 2026.

$22M Series A from Spark and Matrix (Mar 2026)

3 pros · 4 cons

AI quality platform to standardize evals and observability across the organization

Confident AI is an AI quality platform for enterprise teams to standardize evals and observability. It builds and maintains the open-source DeepEval evaluation framework and DeepTeam red-teaming framework. For agents it tests tool selection, arguments and full trajectories, and simulates multi-turn conversations. It blocks releases that fail evals or red-team checks and offers an MCP server and agent skills.

Acquired

Pioneering open-source coding agent (CLI, VS Code, JetBrains), acquired by Cursor in 2026.

Continue Dev, Inc. built Continue, an open-source coding agent available as a CLI, a VS Code extension and a JetBrains plugin. The continue.dev homepage now states that the company has been acquired by Cursor, and that the open-source codebase remains freely available as a foundation for others. The README says the continuedev/continue repository is no longer actively maintained and is read-only. A final 2.0.0 release of the extension, CLI and JetBrains plugin removed anonymous telemetry and pulled out authentication. The code remains Apache-2.0.

A helpful agent and terminal multiplexer for running several coding agents side by side.

Corgi presents itself as "a helpful {agent, terminal} multiplexer" and "friendly helper for your agents": a workspace for running and switching between several terminal coding agents at once. It launched with a Show HN on 23 October 2026 that drew 141 points and 47 comments.

Sovereign AI lab building post-trained coding models and autonomous coding agents.

Cosine builds autonomous software-engineering agents for large existing codebases, built on its own post-trained models, including Genie and the Lumen family (Lumen Scout, Lumen Outpost). Work is assigned through Jira, GitHub or Slack; the agent plans, writes and runs code in a sandbox, tests and validates, then submits PRs for review. It offers CLI, desktop and cloud surfaces with parallel sub-agents and MCP, and deploys to public cloud, single-tenant cloud or air-gapped on-premise environments, with post-training for niche languages such as COBOL and Verilog.

Scale voice AI agents with confidence

Coval is a testing and evaluation platform for voice AI agents. It simulates thousands of realistic caller conversations before launch and evaluates live conversations for quality drift, safety issues and drops in resolution. It also runs regression tests across prompt, model and vendor changes, and has a CLI for running simulations.

Production-like environments with scoped access to real infrastructure for coding agents.

Crafting provides secure agent infrastructure running in the customer's cloud: sandboxes that mirror production, giving agents scoped access to database snapshots, internal services, caches and queues, with network isolation and credentials injected only when needed and never exposed to the model. Crafting for Agents, generally available since March 2026, adds Kubernetes interception, multi-service orchestration, hot-swapping of services and multi-region failover, complementing existing CI/CD. San Francisco-based; raised a $5.5M seed led by Mischief with WndrCo (March 2026). CEO Sumeet Vaidya.

$5.5M seed led by Mischief (Mar 2026)

4 pros · 4 cons

Build and govern MCP servers on company context

Credal is an enterprise platform where teams build and govern MCP servers, with curated context, scoped tools, enforced instructions and parameters, and cost-to-accuracy evals per workflow. It also offers an Agent Builder and an Agent Registry.

Multi-agent orchestration, from open-source framework to enterprise platform.

CrewAI popularized role-based multi-agent teams ('crews') with a simple mental model and ships an enterprise control plane for deploying, monitoring and governing them.

$18M Series A led by Craft Ventures (2024)

3 pros · 4 cons

Cua

Open drivers, cross-OS sandboxes and benchmarks for computer-use agents.

Cua (YC Spring 2025) builds infrastructure for computer-use agents. Cua Spaces is a desktop app that gives agents full desktops: macOS VMs built locally, or Linux images, on the user's own machines or cloud. Cua Driver lets agents inspect and operate native apps and browsers on macOS, Windows and Linux through a CLI, MCP or typed SDKs. The stack also includes Lume (macOS VMs), the Cua SDK and CLI for local sandboxes, CUA-S1 small models for computer-use decisions, and Cua Bench for simulated tasks. The repo is MIT-licensed, while the Spaces app is source-available under FSL-1.1-MIT.

AI code reviews for complex codebases.

cubic is an AI code review tool that checks pull requests against the whole codebase and the team's standards to catch bugs and changes that break code elsewhere. It can auto-approve safe changes, score each PR's merge safety, and pass its feedback to coding agents such as Claude Code, Codex and Cursor for fixes. It also runs scheduled background scans for bugs and vulnerabilities.

The AI-native code editor where agent mode lives.

Cursor rebuilt the IDE around models: tab completion, in-editor chat and agent mode that plans multi-file edits and runs commands. Its rise set the pace for the entire coding-agent market.

~$29B valuation (late 2025)

3 pros · 4 cons

Elastic sandboxes for running AI-generated code.

Daytona provides secure, OCI-compatible sandbox 'computers' that spin up in under 90ms — with snapshots for persistent agent state. In 2026 it moved core development to a private codebase while continuing to run as a commercial platform.

3 pros · 4 cons

Just Use Postgres

DBOS makes workflows durable so applications and AI agents recover from failures, storing workflow state in an existing Postgres database with no separate orchestration server. DBOS Transact is an open-source durable execution library with Python, TypeScript, Go, Rust and Java versions, and integrates with Pydantic AI, LlamaIndex, the OpenAI Agents SDK and the Vercel AI SDK.

AI agents that resolve customer conversations end-to-end.

Decagon builds conversational agents for support and success that take real actions in backend systems, with analytics and guardrails tuned for CX teams.

4 pros · 4 cons

The fastest persistent computer for agents; launched as "Vercel for Agents" connecting any LLM to any MCP tool.

Dedalus Labs launched on Hacker News in August 2025 as "Vercel for Agents", an SDK that connects any LLM to any MCP tool. Its site now positions the company as the "Fastest Persistent Computer for Agents", offering persistent VMs for agent workloads. Y Combinator S25.

Enterprise AI platforms around the Haystack framework.

deepset (Berlin) builds the Haystack framework and an enterprise platform for industrial-grade RAG and agent pipelines — a European anchor of the NLP/agent engineering space since 2018.

3 pros · 3 cons

Customer and Agentic Identity Platform

Descope is an External IAM platform for building identity journeys. It covers users, business customers, partners, AI agents and MCP servers.

A self-hosted worker that turns tracker tickets into pull requests.

DevIntern is a self-hosted daemon that picks up ready tickets from Jira, Linear, GitHub Issues and other trackers, routes each to the right repository, and runs the user's existing coding agent (Claude Code, Codex, Cursor or OpenCode) in isolated worktrees, with optional quality gates and auto-review, before opening a pull request. Review comments go back to the same worker as commits on the same branch. Interactive terminal use is free; unattended automation needs a paid license after a 14-day pilot. The source is public under FSL-1.1-ALv2, and the project works its own backlog in public.

AI testing agent for enterprise unit testing

Diffblue describes itself as an AI testing agent for enterprise unit testing. It automatically generates unit tests for large organizations' codebases.

Open-source platform for agentic workflows and RAG pipelines, cloud or self-hosted.

Dify, developed by LangGenius, is an open-source LLM app development platform combining a visual workflow canvas, RAG pipelines, agent capabilities, model management across hundreds of proprietary and open models, a prompt IDE and observability integrations (Opik, Langfuse, Arize Phoenix). It is offered as Dify Cloud or self-hosted via Docker Compose. The source is under a modified Apache 2.0 license that requires a commercial license for operating multi-tenant services, among other conditions. Its GitHub repository has around 160k stars.

Knowledge platform for teams and agents

Dosu captures what people and agents learn and keeps that knowledge shared and current, so agents use fewer tokens and make fewer mistakes. It syncs with GitHub, GitLab, Slack, Notion and Confluence and stores knowledge as portable markdown. Coding agents connect through the Dosu MCP Server.

A rootless Linux sandbox with gVisor support for running agents safely.

Drop is an open-source, rootless Linux sandbox with gVisor support, built to run coding agents and other untrusted processes without giving them the host. It launched with a Show HN in September 2026 that drew 193 points and 63 comments.

Design, deploy and manage AI assistants for companies.

Dust (Paris) is an enterprise platform for building assistants and agents connected to company knowledge and tools — strong on developer experience and European enterprise adoption.

4 pros · 4 cons

E2B

Open-source sandboxes for running AI-generated code.

E2B gives agents secure, isolated cloud micro-VMs to execute code and commands — the execution runtime layer of the harness, used by many coding and data-analysis agents.

4 pros · 3 cons

Boutique lab crafting tools to harness AI — Pi and Lefos.

Earendil Works 'crafts tools to harness AI': Pi, the minimal, steerable agent toolkit (unified LLM API, agent loop, TUI) that became a favorite base for custom harnesses, and Lefos. Built in the open, shaped by the humans who use them.

Open-source cowork desktop app that runs a local multi-agent workforce.

Eigent is an open-source desktop 'cowork' application for building and running an AI workforce that automates multi-step workflows. Built on the CAMEL-AI open-source framework, it offers a single-agent harness for focused tasks and a multi-agent workforce mode for parallel execution, with MCP and skill integration, built-in browser and terminal toolkits, automations and support for any model, including local ones. It can be self-hosted for free with your own keys; paid cloud plans add task credits, and enterprise features include SSO and access control. The site's legal entity is Eigent UK Ltd.

Managed agent platform: pick the harness, model and tools; Ellipsis runs it

Ellipsis runs coding agents on its own infrastructure and handles VMs, permissions, budgets and logs. Agents are defined in YAML: trigger, harness and model, prompt, repos, hooks, compute and spend limits. Supported harnesses include Claude Code, Codex, OpenCode, Gemini CLI, GitHub Copilot and Cursor. Use cases are software factories, code review and background agents. The company is backed by Y Combinator.

Orchestration for swarms of enterprise agents.

Emergence AI builds an orchestration layer where specialized agents (web, code, data) are dynamically composed into workflows — 'an operating system for agent swarms'.

4 pros · 4 cons

AI agents that design, code and deploy your app

Emergent is an AI development platform that turns plain-language ideas into functional web and mobile apps. Users chat with AI agents that design, code and deploy the application from start to finish.

A learned memory layer for enterprise agents that prepares memories ahead of time to cut token costs.

Engram builds a learned memory layer for enterprise AI agents. Its CEO, Dan Biderman, says the approach can "prepare memories ahead of time instead of pasting them on the fly", reducing what agents must load into context. It came out of stealth in June 2026 with $98M from investors including General Catalyst, Kleiner Perkins, Sequoia and Factory.

$98M at launch (Jun 2026)

3 pros · 3 cons

Model routing, code review and cost insights for engineering agents.

Entelligence builds an engineering 'world model' with products on top: a Model Router that sends each turn to the cheapest model meeting a set quality floor and escalates to frontier models only when needed; Code Review that indexes the codebase and past incidents and runs parallel security, performance and precedent passes on PRs; and Agent Insights that records agent sessions and reports cost per session, retry rate and token waste. A CLI sits in front of Claude Code, Codex and Cursor, with MCP plugins. Founded by Aiswarya Sankar; raised $5M co-led by Mayfield and Correlation Ventures.

$5M co-led by Mayfield and Correlation Ventures

4 pros · 4 cons

A Git forge for agent-human collaboration that versions the reasoning behind code.

Entire, led by former GitHub CEO Thomas Dohmke, is a Git-compatible forge that stores the reasoning behind changes alongside code. Its open-source CLI (first released as Checkpoints) captures agent sessions such as prompts, tools used and token use, initially from Claude Code and Gemini CLI. The forge adds 'Trails' as a pull-request replacement with agent reviews and evals, code-graph plus session memory for agent context, agentic search across code, commits and sessions, and secret redaction. It raised a $60M seed led by Felicis in February 2026.

$60M seed led by Felicis (Feb 2026)

3 pros · 3 cons

Agentic AI runtime security: detects threats and enforces policy on enterprise agents in real time.

Eve Security protects enterprise AI agents at runtime, detecting threats and enforcing policies across platforms such as Databricks, Copilot Studio, AgentCore, Bedrock and Glean. It raised a $4.5M seed round led by Run Ventures in September 2026, bringing its seed total to $7.5M.

$4.5M seed led by Run Ventures (Sep 2026)

4 pros · 4 cons

Raven — the harness of harnesses, built for recursive self-improvement.

EverMind AI builds Raven, an open-source, trusted, persistent and self-evolving multi-agent ecosystem designed for recursive self-improvement (RSI), alongside memory-sparse-attention research for scaling end-to-end memory to 100M tokens.

Exa

Search API purpose-built for AI agents.

Exa (formerly Metaphor) provides neural search and web-crawl APIs tuned for agent consumption — live, cited knowledge for agents instead of stale training data.

4 pros · 4 cons

Droids: AI workers for software teams.

Factory's 'droids' take engineering tickets from issue to pull request, with sandboxed environments and enterprise controls — agents as members of the dev workflow.

4 pros · 4 cons

Background agents for engineering teams

Felan runs background AI agents for software engineering teams. They pick up the recurring work nobody should have to watch, like dependency updates, test fixes, flaky CI and small refactors. Agents start from events, schedules or a person, run on their own, and hand back a result your team reviews. They run in Felan's managed cloud or on your own infrastructure. What makes it different: it's built to keep agent costs low and visible. - No markup on tokens. Use your own API keys or subscriptions, or buy credits at provider rates. - No lock-in. Works with any popular model or provider, including local and private models, so you can use cheap models for simple tasks and top models for hard ones. - Felan Code, our open-source coding agent, is tuned to use fewer tokens without lowering quality. It's the same agent on your laptop and on the platform. - Per-task cost visibility. You see what every agent and task costs. Felan sits next to the local agents your developers already use. It doesn't replace them.

Turn websites into clean, LLM-ready markdown.

Firecrawl (the team formerly known as Mendable, now renamed after the product) crawls, scrapes and structures the web for agents — the de-facto 'eyes' of many research and automation harnesses. Core is open source.

4 pros · 5 cons

Persistent Linux computers for AI agents, with checkpoints and sleep/wake

Sprites is a hosted service from Fly.io that gives AI agents full Linux computers with persistent disks. It takes live whole-disk checkpoints and can restore from them, and idle Sprites sleep and wake when a request reaches their HTTPS URL. Connectors let agents call services like GitHub or Slack without the credential living inside the Sprite, and there are a CLI, a REST API and SDKs for JavaScript, Go, Python and Elixir.

Messaging for the AI era, hosting provider for fmsg protocol

Gives people and programs an address like @user@domain.com which can then message any other fmsg address, much like email. However the fmsg protocols is more efficent, takes care of sender verification and message integrity out-of-the-box. Aiming to be the messaging layer AI agents and people use to collaborate intra and inter org.

Full Linux VMs for AI agents that boot fast, clone live and hibernate.

Freestyle (YC S24) builds VMs for AI agents intended for tasks running hours to weeks. Its site lists fast VM startup, live cloning of running VMs including memory, pause and resume with state intact, default-deny networking with per-VM rules, live preview URLs with custom domains, WireGuard tunnels to private networks, nested virtualization for Docker, and a full kernel with FUSE and eBPF. YC describes VMs that can be snapshotted and hibernated. Founded in 2024 in San Francisco by Benjamin Swerdlow and Jacob Zwang.

Open-source platform for evaluating, tracing and improving AI agents.

Future AGI builds an open-source, self-hostable platform for shipping AI agents that covers the lifecycle in one feedback loop: simulations of edge cases before launch, evaluations, tracing and observability in production, real-time guardrails, an LLM gateway, datasets and optimization. It positions itself as a replacement for stitching together separate tools for evals, observability, guardrails and simulation. The platform is Apache 2.0 licensed, with a free cloud tier alongside self-hosting.

Acquired

Evaluation platform for LLMs and agentic systems.

Galileo (makers of the Luna evaluation models) scores agent runs for hallucination, tool-use errors and task completion — automated quality gates for harness engineering. Cisco acquired Galileo in April 2026 and folded it into Splunk's agent observability; the old rungalileo.io domain has lapsed.

4 pros · 3 cons

Open-source evals, red teaming and test generation for agentic systems.

Giskard builds an open-source (Apache-2.0) Python library for testing and evaluating agentic systems, alongside a commercial AI red-teaming and LLM security platform. Giskard v3 is a modular rewrite for dynamic, multi-turn agent testing: Giskard Checks for evals from simple assertions to LLM-as-judge, and giskard-scan for vulnerability scanning and RAG evaluation, built on giskard-core, giskard-llm and giskard-agents. It can wrap an LLM, a black-box agent or a multi-step pipeline. v2 remains available but is no longer actively maintained.

Acquired

A code change verification agent: risk-rated PR reviews and CI fixes that iterate until the build is green.

Gitar reviews pull requests and rates each change low, medium or high risk, reads CI failures to separate real breakage from flaky tests, and generates fixes; with auto-apply on, it commits them and iterates until CI passes. gitar.ai now redirects to SonarSource, which sells it as Sonar's code change verification agent.

GitHub's AI coding agent, from open issue to merged pull request.

GitHub Copilot spans editor, terminal, desktop and github.com. Its cloud coding agent takes tasks from the Copilot panel, dashboard, the repository Agents tab or scheduled and event-driven automations, and works on a branch in either a GitHub Actions-based environment (custom compute, self-hosted runners, 59-minute session limit) or a multi-repo cloud sandbox, opening one pull request per task. Session logs record its work, rulesets and branch protection still apply, and Agent Merge can watch PRs to address failed workflows or merge conflicts.

GitHub's background agent that takes delegated tasks and opens pull requests.

Copilot coding agent (now part of the broader Copilot in the cloud experience) researches a repository, plans changes, edits files and runs tests and linters in the background, then opens a pull request. Sessions can run in a GitHub Actions-based environment (custom compute, network and setup config, self-hosted runners) or in cloud sandboxes that can span repositories. It is customisable with custom instructions, custom agents and skills, MCP servers and hooks; Actions-based sessions are capped at 59 minutes and each task targets one branch and one PR.

GitLab's platform for orchestrating AI agents and flows across the software lifecycle.

GitLab Duo Agent Platform provides specialized agents for planning, coding, security analysis and analytics that use GitLab context, plus a catalog for managing them. Flows chain agents into multi-step automations, such as diagnosing and fixing a broken CI/CD pipeline. Flows can be triggered by GitLab events, schedules or external tools, and include human approval checkpoints. It is offered to Premium and Ultimate customers on GitLab 18.8 or later.

MCP server registry, inspector and gateway

Glama runs an MCP server registry that it describes as a superset of the official MCP Registry, with every server scanned, scored and testable in the browser. The site lists 112,011 MCP servers, 38,854 connectors and over 1.1M tools indexed.

The durable runtime for your AI agents

Golem runs AI agents and their tools in lightweight WebAssembly VMs and resumes them after crashes, restarts or long waits without repeating completed calls. The core runtime is open source on GitHub, written in Rust.

Google's agentic development platform

Google Antigravity is an agentic development platform with four parts. The App is a command center for running multiple local agents in parallel. The CLI is terminal-first, with background subagents. The IDE is agentic, with an agent manager and artifacts. The Python SDK is for prototyping custom agents on the Antigravity harness.

Google's asynchronous Gemini coding agent that works in a cloud VM and opens PRs.

Jules is Google's autonomous coding agent for tasks such as bug fixes, version bumps, tests and features. Users pick a GitHub repository and branch and write a prompt, or add the 'jules' label to an issue; Jules clones the repo into a cloud VM, drafts a plan with Gemini, shows a diff for approval and then creates a pull request. Plans range from 15 tasks/day with 3 concurrent tasks up to 300 tasks/day with 60 concurrent tasks on the Ultra tier.

The next generation of code review.

Graphite is a code review platform. It runs automated AI reviews on every pull request before merge to catch serious bugs and suggest fixes. Graphite Chat lets teams work with an AI assistant on the PR page to understand changes, resolve CI failures and apply fixes. Cursor Cloud Agents are available in Graphite for building and deploying code from the pull request.

Context layer for AI agents

Graphlit positions itself as the context layer for AI agents, supplying them with context from connected data sources. It maintains an open-source MIT-licensed MCP server in TypeScript.

AI code review agents with full-codebase context.

Greptile reviews and tests pull requests with context from the whole repository. It builds a graph index of files, functions and dependencies, then runs a swarm of parallel agents that assess a change's impact beyond the diff, and learns team standards from reviewers' comments over time. The product line also includes a CLI, security review and TREX runtime validation (beta), framed as an independent check on code from both humans and coding agents. Founded in 2023 and part of YC W24, the San Francisco company raised a $50M Series B led by Menlo Ventures at a $750M valuation in October 2026.

$50M Series B led by Menlo Ventures at $750M valuation (Oct 2026)

4 pros · 4 cons

A complete coding-agent harness — confidence-gated routing, institutional memory.

Grist (grist.lol) is a complete coding-agent harness: every task is routed to the cheapest capable model rung by confidence gates, every verified change is remembered as institutional memory, and inference is billed to your own provider key. npm install -g grist-ai.

The AI Reliability Platform

Guardrails AI provides a guardrails framework for building, governing and scaling production GenAI with any LLM and deployment environment. It offers runtime guardrails that block bad outputs, a Guardrails Hub of validators, and tooling to generate eval datasets that surface agent failures. The open-source guardrails library is Apache-2.0 licensed Python.

Red-teaming and robustness for AI agents.

Haize Labs stress-tests agents with automated adversarial attacks — prompt injection, tool misuse, unsafe actions — and ships an SDK for continuous safety evaluation in CI.

3 pros · 3 cons

AI-native software delivery platform with purpose-built agents across the SDLC.

Harness brands itself 'The Autonomous SDLC Platform'. It runs purpose-built agents across CI/CD, security, runtime protection and cloud cost, including a Software Delivery Agent, a Zero-day Remediation Agent, a Security Testing Agent, a Runtime Protection Agent and a Cost Management Agent, on top of a Harness AI layer and an SDLC knowledge graph. Controls include policy-as-code, an audit trail, human-in-the-loop approval and risk-based autonomy. It is an agents-for-software-delivery company rather than a general-purpose agent harness; a third-party review describes its AI as an add-on to the CI/CD platform. The shared name with the term 'harness' is a coincidence.

The orchestration engine for teams who ship

Hatchet is a platform for orchestrating AI agents, background tasks and mission-critical workflows written as code in TypeScript, Python or Go. It handles retries, dependencies, scheduling, cron and event triggers, and stores the history of every task for debugging and replay. Workers run on your infrastructure with the engine managed or self-hosted; the project is MIT-licensed open source.

Open-source LLM observability and AI gateway.

Helicone is an open-source LLM observability platform — logging, caching, rate limits and evals behind a one-line proxy — with an AI gateway for routing and cost control. A YC company that became the self-hostable counterweight to LangSmith.

~$10M

3 pros · 4 cons

User modeling and psychology for agent memory.

Honcho is a user-context and memory server that models what each user knows and wants, so agents can personalize without teams hand-rolling their own user-modeling stack.

4 pros · 4 cons

Production AI starts with observability

HoneyHive is an agent observability and evaluation platform for enterprises. It traces prompts, tool calls, outputs and handoffs, monitors live traffic, and alerts on scores. It runs offline experiments, builds regression suites from production traces and runs evals in CI. It is OpenTelemetry-native and can be deployed as SaaS, hybrid or fully self-hosted.

Effortless cloud coding agents that carry your local sessions, memories and MCP servers to the cloud.

Hoplite runs coding agents in the cloud and ports a developer's local sessions, memories and MCP servers with them, so work can continue off the laptop. The Y Combinator S26 company pivoted into coding-agent infrastructure from an earlier retail-investing product and launched on Hacker News in August 2026.

Human approval and context engineering for agents.

HumanLayer builds tooling for human-in-the-loop agent workflows and authored '12-Factor Agents' — the widely-cited methodology for building production harnesses. Now focused on agentic code review tooling.

3 pros · 4 cons

The powerful agent for work, built on a shared company brain.

Hyper launched on Hacker News in June 2026 as a shared "company brain" to power agentic development, arguing that "the bottleneck now is that these smart-enough models often lack information about your company". Its site now presents Hyper as "the powerful agent for work". It is part of Y Combinator's Spring 2026 batch.

Cloud browsers, sandboxes and web data APIs for AI agents

Hyperbrowser hosts cloud Chrome sessions that can be controlled with Playwright or Puppeteer, with persistent logins and live inspection. It also offers isolated sandboxes with snapshots and a web API for fetching, crawling and search that returns Markdown, HTML, screenshots or JSON matching a schema. Agent tasks can be run with HyperAgent, Browser Use or computer-use models, and the service ships Node.js and Python SDKs, a CLI and MCP.

An AI on-call agent that drops read-only probes into production to debug live systems.

HyperProbe launched on Hacker News in August 2026 with a way for Cursor and Claude to place read-only probes in production and debug live systems. Its site now leads with an "AI on-call agent" built on the same idea. Y Combinator S26.

IBM's AI coding partner for the codebase, the IDE and the command line.

IBM Bob is an AI coding partner that works inside your codebase, in your IDE and from the command line. It can spawn focused agents and subagents that run tasks in parallel, each with its own context, tools and skills. The site offers downloads, a free trial and pricing.

Agents that use computers — and the sandboxes to train them.

Imbue (formerly Generally Intelligent) builds agents that reason and operate computers autonomously, and open-sourced Sextant, a sandbox environment purpose-built for computer-use agent training and evaluation.

3 pros · 4 cons

Durable functions and workflows for AI agents.

Inngest runs agent logic as durable steps — queueing, retries, concurrency control and human-in-the-loop waits — without operating your own workflow infrastructure.

4 pros · 5 cons

Turns expert workflow knowledge into lasting memory for AI agents.

Interloom captures how experts actually work and turns that workflow knowledge into lasting memory that AI agents can use. It announced a $16.5M seed round led by DN Capital in March 2026.

$16.5M seed led by DN Capital (Mar 2026)

4 pros · 4 cons

The self-hosted governance layer between your AI agents and your business systems, on your rules.

Self-hosted MCP gateway for small and mid-sized companies, delivered as one Docker container. Credentials stay in a built-in vault, permissions are deny-by-default per token and team, every tool call is written to an audit log, and personal data can be pseudonymised deterministically before it reaches the model. Connectors cover REST, upstream MCP, SQL and files, including German ERP and accounting systems. Open core: connector specs, plugins, recipes and deploy files are Apache 2.0.

JetBrains' model-agnostic coding agent for the terminal, IDEs and CI.

Junie is JetBrains' coding agent, offered as a CLI, inside JetBrains IDEs (via Air), as a GitHub Action triggered from issues and pull requests, and in GitLab CI/CD. It supports 10+ models via bring-your-own-key and lets users plan on a strong model and implement on a faster one. Harness features include an advanced plan mode that writes editable, committable plans to .junie/plans, live prompting to steer mid-task, human-in-the-loop approvals with a dynamic allowlist, async remote-control runs monitored from a web app, and persistent custom guidelines and skills.

Find and fix what your agents get wrong

Judgment Labs makes a monitoring and improvement platform for long-horizon AI agents. It traces agent runs including subagents and tool calls, monitors custom signals with code checks or agentic judges, and surfaces recurring production failures. It also runs offline regression evals, and offers a Slack agent and an MCP server for use with coding agents. It can run in Judgment's cloud or in the customer's own AWS or GCP account.

Plano: AI-native proxy and data plane for agentic apps.

Katanemo develops Plano, an open-source (Apache-2.0) AI-native proxy server and data plane for agentic applications, built on Envoy and written in Rust. Plano moves delivery concerns out of application code into an out-of-process data plane: low-latency routing and orchestration between agents, LLM routing by model name, alias or preferences, zero-code capture of 'Agentic Signals' plus OpenTelemetry traces and metrics, and filter chains for jailbreak protection, moderation and memory hooks. It works with any language or agent framework.

A full-stack cloud with forkable VM snapshots and global SQLite, deployable over SSH or Git.

Kedge is a cloud for deploying websites, functions and container images over SSH or Git; its Show HN in July 2026 (141 points) pitched forkable VM snapshots and global SQLite, features that suit agents which need disposable, branchable machines.

Browser infrastructure for web agents and automations

Kernel provides isolated cloud Chromium sessions controlled through an API, SDK or CLI, using Playwright, CDP, WebDriver BiDi or computer use. It supports persistent profiles, managed auth that keeps credentials and MFA codes out of the agent conversation, and stealth mode with captcha solving and proxies. It also has live view, session replay, a hosted MCP server, and browser pools for latency-sensitive work.

Runtime auth for agents

Keycard provides runtime authorization for AI agents. Each agent gets its own identity, authority to act, and access to resources under the owner's policy.

Moonshot AI's hosted, always-on OpenClaw deployment on kimi.com.

Kimi Claw is Moonshot AI's managed version of the OpenClaw personal-agent harness, announced in February 2026. It deploys an OpenClaw assistant to the cloud with one click to run 24/7, auto-configured with Kimi models and skills; Dataconomy reports it runs on Kimi K2.5 with 40GB of cloud storage and access to ClawHub's 5,000+ community skills. A 'Bring Your Own Claw' option links existing self-hosted OpenClaw setups, and the page also offers local desktop and Android deployment. The kimi.com page marks it as beta.

AWS's spec-driven agentic IDE, CLI and autonomous cloud agent.

Kiro is an agentic AI coding tool built and operated by AWS that helps developers plan, build and review software. Its spec-driven workflow turns prompts into requirements, architectural designs and sequenced tasks before code is written, then implements them with parallel agents. Hooks run background tasks such as adding tests or updating docs when code changes, and steering files carry project context across surfaces. Kiro ships as an IDE, a CLI that can run headless in CI/CD, Kiro Web with an autonomous mode running in isolated cloud sandboxes, a mobile app, and Kiro Crew, an open-source workspace.

MCP integration platform and live environments for AI agents

Klavis AI's open-source platform provides MCP integrations so agents can use tools reliably at scale. The company also builds live environments and training data for agents, including Dockerized long-horizon coding tasks and tool-use data across 600+ tools and SaaS apps. It is backed by Y Combinator.

Open-source AI code review agent (Kody)

Kodus presents itself as the open source alternative to CodeRabbit. Its reviewer, Kody, reads diffs and posts inline comments on GitHub, GitLab, Bitbucket, Azure DevOps and Forgejo/Gitea. The core is AGPL, a commercial license is available for enterprise use, and teams can self-host it with Docker Compose or Helm.

AI design engineer for websites and product UIs

Kombai is an AI design-engineer agent that builds websites and product UIs fitting an existing product. It reuses design systems, tokens and components from repos, npm or Storybook, offers a design mode on a canvas and a code mode producing production code, and routes across 50+ models. It runs as a desktop app or VS Code/Cursor extension and can work alongside Claude Code or Codex.

Acquired

AI code review agent for pull requests

Korbit AI is an AI code review tool for GitHub, GitLab and Bitbucket that gives real-time, actionable feedback in every pull request. It also generates PR descriptions, review summaries and team reports. Boost Security announced in 2026 that it had acquired Korbit.ai as part of its SDLC defense platform.

Open-source observability platform purpose-built for AI agents.

Laminar (YC S24) is an open-source observability platform for AI agents. Its OpenTelemetry-native SDK auto-traces Vercel AI SDK, Browser Use, Stagehand, LangChain, OpenAI, Anthropic and Gemini. Signals let teams describe behaviors in plain English, such as an agent stuck in a loop, and get Slack alerts when they occur. It also offers an evals SDK and CLI for local or CI runs, SQL-queryable traces via MCP and CLI so coding agents can debug from traces, custom dashboards, and dataset annotation. The backend is written in Rust and can be self-hosted with Docker Compose.

AI Cloud in one line of code

Langbase is a serverless platform for building, deploying and scaling AI agents, which are exposed as APIs (pipes) and can connect to 600+ LLMs, tools and MCP servers. It provides a managed memory/RAG service, and multi-step multi-agent workflows with retries, timeouts and step-level traces. It also offers Studio for versioning agents, plus tracing, cost tracking and evals. Its site also promotes Command Code, a coding agent.

Acquired

Open-source LLM engineering platform: tracing, prompts, evals.

Langfuse is an open-core platform for observing and improving LLM apps — trace-level debugging, prompt management, datasets and evals. Self-hostable, with a cloud offering. Acquired by ClickHouse in January 2026 as part of its push into AI observability; the open-source core remains actively maintained.

Acquired by ClickHouse (Jan 2026)

4 pros · 4 cons

Secure AI adoption at enterprise scale

Lasso is an AI security platform covering AI discovery and inventory, AI agent security, AI governance and AI usage control. Its use cases include AI agent governance, AI gateway security, MCP and prompt injection, and AI coding assistants. It has announced CPU-based guardrails.

Open-source agent observability that dispatches coding agents to fix failures.

Latitude (Latitude Data S.L.) builds an open-source, MIT-licensed platform for monitoring AI agents in production. One line of telemetry captures multi-turn sessions, tool calls and full execution paths, with OTEL compatibility. Failing traces are auto-grouped into tracked signals and issues with alerts. Latitude then dispatches a coding agent such as Claude Code or Cursor with sample traces to write a fix and open a PR, and replays fixes against the real failing traces as regression tests. Everything is also exposed through an MCP server and CLI. Hosted data stays on European servers.

Stateful agents with memory, from the team behind MemGPT.

Letta (formerly MemGPT, out of UC Berkeley) builds an agent server with built-in memory, context management and state — letting agents persist, learn and be edited like files rather than living in a chat window.

$10M seed led by Felicis (2024)

4 pros · 3 cons

AI employees that run your workflows.

Lindy lets anyone build 'AI employees' — email triage, meeting prep, support — with an event-driven agent runtime and a large template library.

4 pros · 4 cons

The data framework for connecting agents to your data.

LlamaIndex provides ingestion, indexing and retrieval primitives that give agents grounded context over enterprise data — a foundational piece of the retrieval side of the harness stack.

4 pros · 3 cons

Agent collaboration platform that hires, schedules and reports on a team of AI agents.

LobeHub, formerly the LobeChat project, describes itself as an agent collaboration platform for work. It casts itself as a 'Chief Agent Operator' that runs agents around the clock: it hires and schedules them and reports on the work. Its focus is persistent agent teammates and group collaboration with shared context. The source is on GitHub (repo created 2023) under the LobeHub Community License. That license is Apache 2.0 with extra conditions, and distributing derivative works needs a commercial licence. Users can self-host with Docker or Vercel, or use the hosted LobeHub Cloud at app.lobehub.com.

Custom models and harnesses trained on your data; launched to debug and evaluate agents in production.

Lucidic launched on Hacker News in July 2025 to debug, test and evaluate AI agents in production (116 points). Its site now reads "Custom Models & Harnesses Trained on Your Data", a move from observing agents to training the models and harnesses behind them. Y Combinator W25.

Acquired

Enterprise MCP gateway and AI control plane

Lunar.dev offers an enterprise MCP gateway, an AI gateway and agent governance to secure and govern AI applications and agents. Its open-source repo describes an agent-native MCP gateway for governance and security. The site says Boomi has completed its acquisition of Lunar.dev.

Command your own software factory.

Macroscope offers Murmur, in beta, which orchestrates cloud agents that write code and verify their work in isolated sandboxes. Its Code Review feature checks each change for correctness, security, tests and regressions before merge. Its Status feature shows what changed, why, and what needs attention across a team's repositories.

An open-source inference engine for agents that tunes its kernels to your hardware.

Magnitude is an open-source local inference engine that tunes its kernels on the user's hardware to find the fastest way to run each open model, which it says is over 2x faster than Ollama (company claim). It launched on Hacker News in September 2026 with 194 points. Y Combinator S25.

Fullstack MCP framework and cloud for MCP apps and servers.

Manufact is the company behind mcp-use, an open-source (MIT) TypeScript framework for building, testing and shipping MCP servers, MCP Apps for ChatGPT and Claude connectors. The framework offers Zod-typed tools, React 'Views' bound to tools, a built-in Inspector and an agent-first, headless CLI for scaffolding, invoking and deploying servers. Manufact's site positions it as 'the cloud for production MCP', helping teams get onto ChatGPT and Claude plugin stores, see how agents use their servers and test releases across clients and harnesses.

Acquired

General-purpose cloud agent with its own computer, acquired by Meta.

Manus is a general-purpose AI agent launched in 2025 by Butterfly Effect, which was founded in China before moving to Singapore. Subscribers use it for research, coding and other long tasks, and its current site lists building slides, websites, games and video, Wide Research, a Slack integration and an API, across web, desktop and mobile apps. Meta announced its acquisition of Manus on December 29, 2025; terms were undisclosed (the WSJ reported over $2B), and Manus said it would keep operating its subscriptions.

Acquired by Meta (Dec 2025

2 pros · 4 cons

Run coding agents without babysitting them

MartinLoop is a runtime layer that wraps AI coding agents such as Claude Code, Codex and Gemini CLI. It lets teams hand agents larger software jobs with spend limits, required checks before results are accepted, and a record of each run.

The TypeScript-first company behind the Mastra agent framework.

Mastra builds the eponymous open-source TypeScript framework — workflows, tools, RAG, evals and a local dev playground — and is becoming the default 'agent backend' choice for JS/TS teams.

4 pros · 4 cons

Bifrost: open-source LLM, MCP and agent gateway.

Maxim AI builds Bifrost, an open-source (Apache-2.0) AI gateway written in Go that acts as an LLM gateway, MCP gateway and agent gateway. It exposes one OpenAI-compatible API across providers including OpenAI, Anthropic, AWS Bedrock, Google Vertex, Azure, Mistral, Groq and Ollama, with automatic failover, load balancing, semantic caching, virtual keys, budgets and an MCP 'Code Mode'. The project claims under 200 microseconds of overhead at 5k RPS. Enterprise deployments add clustering, guardrails, adaptive load balancing and private networking.

MCP gateway for security, deployment and observability

MCP Manager by Usercentrics is an MCP gateway for AI governance. It provides guardrails, RBAC, logging and visibility for companies in regulated industries. Features include an MCP registry, logs and observability, and PII filtering.

The memory layer for AI apps and agents.

Mem0 (from the EmbedChain team) provides a managed + open-source memory layer that extracts, stores and retrieves user and entity memories across sessions — a standard building block for personalization.

3 pros · 4 cons

Agent-native memory infrastructure that turns agent execution into persistent state.

Memori Labs builds Memori, an LLM-agnostic memory layer that turns agent execution traces and conversations into structured, persistent state. The open-source SDKs (pip `memori`, npm `@memorilabs/memori`) register an existing LLM client and attribute memories to entities and processes so later runs recall them automatically; a Memori Cloud service and a gateway on the inference path are offered commercially. The site emphasizes governance (access-controlled memory pools, audit logs) and deployment across managed cloud, single-tenant cloud, VPC and on-prem. The repo is Apache-2.0.

Open-source MCP and skills gateway

Metorial gives an organization one MCP URL behind its SSO, with 1,000+ integrations, role-based tool access, audit logs and shared skills in Claude, ChatGPT and Cursor. Its GitHub repo describes it as connecting any AI model to 1200+ integrations via MCP, CLI or API.

MCP & LLM Gateway for the Agentic Enterprise

Mintive, formerly MintMCP, is an MCP and LLM gateway. It gives every agent its own identity and governs access to models and tools by role.

Minimal CLI coding agent by Mistral

Mistral Vibe combines Mistral's work chat assistant (formerly Le Chat) with a coding agent. The agent can be installed as a CLI, used inside VS Code, JetBrains and Zed via ACP, or run as remote agents. The CLI agent is open source.

Serverless cloud built for AI and agentic compute.

Modal lets teams run inference, fine-tuning and arbitrary agent workloads as serverless Python without managing infrastructure — a common substrate for harness backends at scale.

4 pros · 4 cons

The deterministic engine for agents to make estate-wide code changes.

Moderne offers agent tools, including Trigrep and Prethink, and a CLI that can direct GitHub Copilot to run changes, such as a Java 25 upgrade, across a set of repositories. Each agent-driven 'factory' runs a recipe and observes its effect across every repository in the set. Build or test failures in one repository feed back to refine the recipe, so it can make the change safely in repositories with weaker test coverage.

AI QA engineer that tests your app end to end

Momentic's agent Mo tests apps end to end from plain-English flows stored as YAML in the repo. It runs locally and in CI as a merge check. Coding agents can author and verify tests through MCP. It triages failures as app bugs, flakiness or environment issues, and proposes repairs to tests when the UI changes.

Fast specialized models for coding agents: apply, search, compaction and routing.

Morph (YC S23) provides a model API for coding agents. Fast Apply models merge LLM-generated edits into files (the site cites 10,500 tok/s), WarpGrep is a code-search subagent, and Compact compresses context while keeping original bytes; it also serves open-weight agent-loop models and a router. Access is via an OpenAI-compatible API, TypeScript and Python SDKs, OpenRouter, an MCP server, and CLI/skills for Claude Code, Cursor, Codex and OpenCode, with self-hosting for enterprise. YC lists founder Tejas Bhakta and Columbus, OH.

Connect your product and AI agents to 1,000+ APIs

Nango is integration infrastructure for connecting products and AI agents to 1,000+ APIs. It is used to build, run and maintain integrations in code. The open-source repo is written in TypeScript.

Make Your Website Actionable For Agents

nekuda makes websites actionable for agents using WebMCP. Sites can build full tools with a coding agent or start from a scan of the site, and get full visibility into agent activity.

Nia

A context layer for coding agents that indexes code and documentation for Cursor, Claude Code and MCP clients.

Nia indexes codebases and documentation and serves that context to coding agents such as Cursor and Claude Code, and to other MCP clients, so agents answer from the real source rather than guesses. The Y Combinator S25 company launched on Hacker News in December 2025 with 131 points.

Agent security and governance platform

Noma offers a unified agent control plane to secure agents on SaaS platforms, endpoints and cloud. Capabilities include agent runtime security, MCP security, agentic identity, AI asset discovery and AI supply chain security.

Runtime platform for AI agents, their sandboxes and the software they build

Northflank runs agent workloads in isolated microVM sandboxes and can host coding agents such as Codex, Claude Code, OpenCode and Pi remotely with persistent storage. It also runs services, managed databases, jobs and GPUs, with preview environments per PR and pipelines. It runs on Northflank Cloud or in a customer's own cloud or bare metal, and the site states a registered office in London.

Browser platform and framework for AI web agents

Notte provides cloud browser sessions with anti-detection, residential proxies and CAPTCHA solving, plus browser agents that navigate sites from natural-language tasks. It can deploy browser automations as serverless functions with cron scheduling, and offers personas and secret vaults so credentials stay away from the LLM. It has an MCP server, a CLI and an agent skill for coding agents. The site says it is backed by Y Combinator (W25).

Open AI lab behind the Hermes models and Hermes Agent.

Nous Research, known for open-weight Hermes LLMs and decentralized training, ships hermes-agent — a self-improving, always-on personal agent that turns experience into reusable skills and keeps a persistent user model, lean enough to run on a $5 VPS.

Venture-backed (2024–25 rounds)

Open-source AI governance platform and MCP gateway

Obot's GitHub repo describes it as a 'Complete AI Governance Platform from Obot AI'. The open-source project is written in Go under the MIT license and is actively maintained.

Simulate real users. Ship reliable voice and text agents.

Okareo tests and monitors voice, text and headless AI agents using synthetic users called Drivers. Drivers have a personality, context and goal, and hold long multi-turn conversations to surface edge cases. Simulations and evaluations run on each commit or pull request, and production failures can be turned into new test scenarios. It also lists agent monitoring and an MCP integration.

Acquired

Background software-engineering agents in governed, VPC-deployable cloud environments.

Ona, the company formerly known as Gitpod (Gitpod GmbH), runs AI engineering agents in full cloud development environments that can be deployed in the customer's VPC. Background agents take a task and return a pull request; automations trigger agent fleets from PRs, schedules or webhooks for work like CVE remediation and code migration. Runtime controls include audit trails, scoped credentials and kernel-level policy enforcement, and teams can run Ona, Codex and other agents under shared policies. OpenAI announced on June 11, 2026 that it is acquiring Ona, with the team joining Codex.

Acquisition by OpenAI announced June 11, 2026 (terms undisclosed)

ART: open-source reinforcement learning trainer for multi-step agents.

OpenPipe maintains ART (Agent Reinforcement Trainer), an open-source (Apache-2.0) RL framework that lets LLM agents learn from experience by integrating GRPO into any Python application. Project news covers a LangGraph integration, MCP-RL for training models to use MCP server tools, AutoRL for zero-data training and RULER for LLM-judged rewards. The README also promotes W&B Training (Serverless RL), a managed service for running ART training and inference. Supported open models include Qwen, GPT-OSS and Llama.

One API for every model — routing, pricing and fallbacks.

OpenRouter is the unified API for frontier and open models: one endpoint, per-token billing across providers, automatic fallbacks and routing by price, latency or capability. Bootstrapped and profitable, it became the default way provider-agnostic harnesses reach every model.

4 pros · 4 cons

Secure your AI

Operant AI offers real-time discovery, detection and defense for AI, agents and MCP. Products include Agent Protector, an MCP Gateway with registries and allow/deny lists, a Semantic Firewall that allows/blocks/redacts agent actions, and an Endpoint Protector that monitors coding agents such as Claude Code and Cursor on employee devices.

A context- and intent-aware AI software factory with guardrailed agents.

Opsera sells an AI-driven software delivery platform; its Forge product, launched in April 2026, is a governed 'software factory' that converts intent into machine-readable specifications before code is written, reverse-engineers legacy COBOL, .NET and Java systems into specs, and keeps a persistent context layer to reduce drift. Humans supervise agents through auditable work orders, agentic actions are mapped to corporate policies, and a ForgeScore scorecard rates output across eight dimensions. Opsera also offers IDE AppSec agents, AI-spend analytics and policy-as-code DevOps.

Stably AI's open-source environment for running fleets of coding agents in parallel worktrees.

Orca is an MIT-licensed agent development environment from Stably AI, a Y Combinator-backed San Francisco company. It runs coding agents such as Claude Code, Codex, OpenCode, Pi, Gemini and Cursor CLI side by side, with about 27 agents supported. Each task gets its own isolated git worktree, so one prompt can fan out to several agents and the best result can be merged. Other features are GPU-rendered terminal splits, a per-worktree Chromium 'Design Mode', GitHub and Linear integration, SSH remote runtimes, an Orca CLI that agents can drive, and iOS/Android companion apps.

A control plane for an AI software factory: governed, stateful agent workflows across the software lifecycle.

Overcut runs AI coding agents as governed, stateful workflows for code review, CI fixes, ticket triage and design-to-PR work. Teams build agents on any model (Claude, Cursor, GitHub Copilot or OpenAI agents), connect GitHub, GitLab, Bitbucket, Azure DevOps, Jira, Linear and Slack, and add approval gates, RBAC, audit trails and per-run sandboxes. It runs as SaaS, in a customer VPC or on-prem, and Gartner's Innovation Insight: AI Software Factories report (September 2026) names it as an example of a horizontal agentic software engineering platform.

An engineering 'superharness' that runs coding agents as one governed team.

Palgu describes itself as an engineering superharness that takes a task from intent to a proven, governed code change. It runs coding agents such as Claude Code, Codex, Cursor and aider as one team and wraps them in eight layers: intent, architecture, knowledge, coordination, evidence, recovery, governance and learning. Its products are the palgu CLI and the Palgu Studio desktop app for macOS and Windows, plus open-source Maggy (a single-developer harness), the Mixture of Models mod for Claude Code and other agents, and AWG, a shared work graph for coding agents. Features include a code graph with blast-radius mapping, a five-seat review council of models other than the author, an audit ledger, BYOK and a self-hosted gateway.

6 pros · 4 cons

Open-source orchestration for running a company of AI agents: org charts, goals, budgets and governance.

Paperclip, from Paperclip Labs, Inc., is an MIT-licensed, self-hostable Node.js server and React UI that orchestrates teams of AI agents around business goals. It pitches itself as 'If OpenClaw is an employee, Paperclip is the company'. Users bring their own agents and harnesses (OpenClaw, Codex, Claude, Cursor and others) and choose them per agent. Paperclip adds org charts and reporting lines, goal-linked tasks, scheduled heartbeats and routines, cost budgets with threshold pauses, approval stages, shared skills, connector access control and multi-org isolation.

Agentic IDE for running Claude Code, Codex and other CLI agents in parallel worktrees.

Paradev (formerly Superset; unrelated to Apache Superset) is a desktop workspace for running CLI coding agents such as Claude Code and Codex in parallel, each in its own git worktree with its own branch, terminals and per-workspace browser preview. It includes a diff viewer, PR review that sends selected lines back to an agent, a Design mode for UI changes, and shared Pages for team review. Users bring their own agent subscriptions and the app does not proxy model API calls. Source is public under the Elastic License 2.0 (source-available, not OSI open source); macOS is the primary platform. Its site announces an $11.5M seed.

$11.5M seed (per paradev.com)

4 pros · 5 cons

Web infrastructure for AI to search, extract, monitor and reason over the world's information

Parallel offers Search, Extract, Task (deep research), FindAll (entity datasets), Monitor and Responses APIs, built on its own Parallel Web Index. Outputs can include citations, reasoning traces and calibrated confidence scores through its Basis framework. It ships Python and TypeScript SDKs and a Parallel Search MCP server, and works with LangChain and the Vercel AI SDK.

Automated evaluation and guardrails for LLM systems.

Patronus AI scores outputs against policy and business criteria in CI and production, catching regressions before users do — 'unit tests for AI'.

4 pros · 3 cons

Securing the agentic workforce

Pillar Security provides AI discovery and posture management, AI red teaming, runtime protection, and compliance and monitoring for AI applications and agents. It publishes the SAIL framework, a practical guide to securing AI agents.

Build your own AI software factory.

Port is a platform for building, governing and operating an agentic software development lifecycle. It discovers agents, MCP connectors and skills across an organization so teams can enforce their standards. Agents handle tasks such as provisioning infrastructure, resolving tickets and fixing pipeline failures, under governance controls with humans kept in the loop.

AI gateway with guardrails for production agents.

Portkey builds an AI gateway that routes across 1,600+ models with integrated guardrails — caching, fallbacks, load balancing, rate limiting and observability — the control plane between a harness and every model provider. Strong presence in the APAC market.

~$10M

4 pros · 4 cons

A context graph of your codebase and SDLC for AI agents.

Potpie turns a codebase and its software-delivery lifecycle into a living context graph for AI agents, indexing code structure, decisions, source history, team knowledge and workflows so agents can answer questions, plan changes, debug failures and write code with project-specific context. The open-source CLI (Apache-2.0, about 5.8k GitHub stars) runs a setup wizard that provisions a local daemon and agent skills and plugs into the user's chosen coding harness, such as OpenAI Codex. Its enterprise offering runs agents inside the customer's VPC and adds spec-driven development; Potpie says it was one of 20 picks for Google's 2026 Accelerator.

Workflow orchestration, now stewarding FastMCP.

Prefect builds dataflow and workflow orchestration infrastructure, and in 2025 took stewardship of FastMCP — the most popular way to author MCP servers. A data-platform company moving squarely into the agent harness layer.

4 pros · 4 cons

Decentralized AI lab — Prime Agent, the self-improving RLM agent.

PrimeIntellect, the lab known for decentralized training and the INTELLECT model series, open-sourced Prime Agent: a recursive-language-model agent for coding and long-running autonomous tasks whose harness famously took Opus 5 from ~30% to 95.5% on ARC-AGI-3 — evidence that the harness can matter more than the model.

Venture-backed (seed + Series A, 2024–25)

Acquired

GenAI and agent security, from SentinelOne

Prompt Security, now branded 'Prompt Security from SentinelOne', provides AI security with guardrails for sensitive data, secrets and prompt injection. It publishes open-source tools including ClawSec (securing agent skills for OpenClaw, NanoClaw and Hermes agents), OneClaw (tracking OpenClaw deployments) and ps-fuzz (testing system prompt resilience).

Pulumi's AI infrastructure agent

Pulumi Neo is an AI agent that provisions, governs and optimizes cloud infrastructure, covering provisioning, updates, debugging and cost analysis. It enforces existing Pulumi policies (RBAC, compliance, security standards) and supports configurable autonomy with human-in-the-loop approval workflows. Every action is previewed, logged and reversible.

Type-safe Python infrastructure — now including agents.

Pydantic builds the data-validation library underlying most of the Python AI ecosystem, plus Logfire (observability) and Pydantic AI (typed agent framework) — the type-safety layer of the harness.

4 pros · 4 cons

QM

A multiplayer agent harness for work.

QM gives every person and shared room a durable, isolated agent workspace while keeping administration, permissions and background work centralized — docs cover models and harnesses, sandboxes, apps and skills, and Slack/web surfaces. Its open-source runtime (open-multi-agent) ships durable approvals and verifiable run records.

An IDE with coding agents and deep codebase awareness

Qoder is an agentic coding platform with an IDE, desktop and mobile apps, a JetBrains plugin and a terminal CLI. It supports delegated autonomous programming with plan and approval controls, generates technical design specs, and keeps a Repo Wiki of project structure. A China site is at qoder.cn.

Multi-agent code review and quality governance for AI-written code.

Qodo builds a code review and governance layer for agent-written software. Its platform runs multiple specialized review agents on pull requests, with a coordinating layer that filters and prioritizes findings, plus cross-repo review, a software map of service dependencies, a codebase knowledge base and an 'Agentic Toolbox' CLI that gives coding agents such as Codex, Claude and Kiro a quality counterpart inside their loop. In March 2026 it raised a $70M Series B led by Qumra Capital, bringing total funding to $120M.

$70M Series B led by Qumra Capital (Mar 2026)

3 pros · 5 cons

Detect agent issues. Prevent them from happening.

Raindrop monitors AI agents in production, finds silent failures and tests fixes before they ship. It traces every message, tool call and decision, groups recurring failures into issues, and runs a Triage Agent that investigates them in Slack, on the web or through MCP. It also runs simulations of agent changes in CI and offers Workshop, an open-source local tool for replaying and debugging agents.

Slate, a swarm-native coding agent for long-running engineering work.

Random Labs (YC S24) builds Slate, a long-running autonomous coding agent. Slate V1, launched in March 2026, uses a 'Thread Weaving' design in which a central orchestration thread dispatches parallel worker threads through a TypeScript-based DSL; workers return compressed 'episodes' rather than full transcripts, and dynamic pruning manages context in large codebases. Different models can be assigned to orchestration and execution roles, and the CLI uses usage-based credits. Co-founded in 2024 by Kiran and Mihir Chintawar in San Francisco.

Build, deploy and publish agent apps as MCPs.

Recurse makes it easy to build an agent app, deploy it, and publish it as an MCP server usable from Claude, Codex or any MCP client — hosting handled, started from your coding agent via a SKILL.md. A bridge between harness skills and deployable agent infrastructure.

Self-improving coding agents that recover and verify.

Relay is a coding-agent harness where every run teaches the next: a queue with one worktree per task, models you already pay for (Claude, ChatGPT, Copilot or API keys), tests re-run by Relay, and lessons kept for the next lap. Free, local-first, BYOM.

Build an AI workforce that runs itself.

Relevance AI (Sydney/SF) provides a platform for building, orchestrating and monitoring teams of work agents — 'tools' and 'teams' primitives for non-developers.

3 pros · 4 cons

Agent: build and ship software from a prompt, in the browser.

Replit Agent turned the long-standing browser IDE into one of the first mass-market 'describe it and it gets built' agent products, with cloud workspaces and databases attached.

3 pros · 4 cons

AI SRE that goes on-call and resolves incidents

Resolve AI is an AI SRE platform whose agents go on-call on a team's behalf. They triage and mitigate alerts, identify root causes during incidents, fix recurring issues and run custom production automation, escalating to engineers when needed.

The durable runtime for agents and backends.

Restate is a lightweight durable-execution runtime purpose-built for AI agents and workflows — journaling every step so invocations survive crashes, retries are deterministic, and long-running agent tasks run reliably. Raised a $20M Series A in September 2026 to 'define the infrastructure layer for AI agents'.

$27M total · $20M Series A (Sep 2026)

4 pros · 4 cons

Agentic software delivery governed by approval thresholds and audit logs.

Revolte describes itself as an agentic platform for software engineering in which AI agents handle development, testing, deployment and production operations while engineers approve outcomes. Approval thresholds can be set per workflow, team or risk level so consequential decisions route to a human first, and every agent action is logged for audit. It integrates with existing stacks such as AWS, Google Cloud, Datadog, Jira, Linear and ServiceNow. The site footer names Poddle.io Ltd; performance figures on the site are vendor-reported.

AI writes code. Riza runs it.

Riza is an API for running code written by an LLM in isolation from production systems and secrets, with per-run control over environment variables and network access. The site says execution starts in under 10ms with no sandbox startup. It offers Python, TypeScript and Go SDKs, lets agents write reusable tools, and can be self-hosted on Kubernetes.

Devbox sandboxes and agent infrastructure for AI coding agents.

Runloop Devboxes are VM-isolated sandboxes where AI agents build, run and test code, starting in seconds and supporting snapshot, suspend and resume for long-running work. Blueprints define team-shared images; network policies restrict egress, while Agent Gateways and an MCP Hub connect agents to LLM APIs and MCP servers without exposing real credentials. The platform also lists measurement for evaluating agent performance, multi-agent coordination, Reflex debugging/observability and VPC deployment on AWS, GCP or Azure, with Python and TypeScript SDKs and a CLI.

The execution layer that 'parents' AI agents.

Runta builds an agent execution layer with sandboxes, spending caps and audit trails — a 'parenting' layer for agents that directly addresses the containment failures exposed by prompt-injection incidents. a16z called it 'core systems software rebuilt for agents'; $20M seed at a $100M+ valuation (Jul 2026).

$20M seed (Jul 2026)

3 pros · 3 cons

The AI agent harness for payment teams

Runtime is a platform for building and running AI agents across payments, risk, compliance and support workflows. Agents can run in the customer's own cloud, and sensitive steps such as releasing funds or sending replies wait for human approval. The footer says it is built in San Francisco.

The hivemind for human–agent teams: shared team memory that gives coding agents team context.

SageOx builds team memory for human and AI work, describing itself as "the hivemind for human–agent teams". Its tools give coding agents the context a team has built up. The Seattle company raised a $15M seed round led by Canaan in May 2026.

$15M seed led by Canaan (May 2026)

4 pros · 4 cons

The simulation platform for agent self-improvement

Scorecard is a platform for testing, evaluating and improving AI agents. It runs agents through large sets of realistic scenarios before release, and provides a metric library, prompt versioning and a playground. It also supports deploying agents and detecting real-world usage problems.

The Enterprise AI Agent Company

Sema4.ai builds an enterprise AI agent platform. Its GitHub organization ships a VS Code extension for developing AI Actions and automations in Python and infrastructure assets for self-hosted Sema4.ai deployments.

Observability for AI agents: catch drift and silent regressions across sessions and tool calls.

Sentrial monitors AI agents in production for loops, hallucinations, tool misuse and silent regressions across every session, tool call and model interaction ("Catch agent drift before your users do"). It launched on Hacker News in March 2026. Y Combinator W26.

Sentry's AI debugging agent

Seer uses Sentry telemetry to automatically investigate new issues and find root causes. Autofix traces the cause through the codebase and drafts a fix. AI Code Review checks pull requests against production errors and traces. Fixes can be handed to Claude, Copilot or Cursor, and investigations can be run from Slack. It costs $40 per active contributor per month on paid plans.

Computer-use agents: the open-source Agent S framework, the Sai agent and the Simular API.

Simular, founded by former DeepMind scientists, builds computer-use agents that operate desktop and web apps by looking at the screen and clicking, typing and scrolling. Its open-source Agent S framework runs on macOS, Windows and Linux; the project reports Agent S3 as the first to surpass human performance on OSWorld (72.60%) and released Agent S4 in October 2026. Commercially it offers Sai, a production computer-use agent, and the Simular API for giving agents autonomous cloud computers. In December 2025 it announced a $21.5M Series A led by Felicis, with NVentures and South Park Commons participating.

$21.5M Series A led by Felicis (Dec 2025)

3 pros · 4 cons

A ticket-taking coding agent that runs on your Claude or ChatGPT subscription.

Sinatra (sinatra.dev) is a coding agent that takes tickets. Users assign it a Linear issue, label a GitHub issue, mention it in Slack or start a task from the terminal. It checks out the repo in its own sandbox, makes the change, runs the tests with two tries to fix failures, and opens a draft PR with a summary and its own review notes, pushing fixes when reviewers comment. Models run on the user's Claude or ChatGPT subscription or an API key, and a committed .sinatrarc.json can pin the test command, model, engine or reasoning effort. The free tier runs one agent at a time; Pro ($20/month) runs three.

Searches past sessions across coding agents, so skills and memories are not locked into one provider.

Skillsync started from a lock-in complaint: "We get locked into a single provider and their agent as we invest in skills and memories over time." It launched in September 2026 to make sessions portable across coding agents, and its site now leads with fast search over past sessions across agents ("search so fast it feels like memory"). Y Combinator W26.

Access. Identity. Checkout. The Agent Trust Stack.

Skyfire gives agents verified identity and payment credentials that work across the open internet. It offers a Know Your Agent (KYA) credential and an agentic wallet for card and stablecoin payments with recorded user approvals.

A registry and toolchain for MCP servers.

Smithery indexes and helps deploy MCP (Model Context Protocol) servers — a distribution layer for the tool ecosystem agents connect to.

3 pros · 4 cons

The context layer for your entire codebase

Sourcegraph provides code search and code-graph context for developers and AI coding agents, including an MCP server for 'code graph knowledge for agents' and Deep Search with citation-backed answers. Agentic Batch Changes is described as an agent for code changes at scale. The company cites its own CodeScaleBench report for improvements in agent speed and token use.

Perfectionist AI code review.

Sourcery is an AI code reviewer for GitHub and GitLab that reviews each pull request within minutes, posting a summary, review comments and fixes that can be committed directly. It can auto-approve low-risk changes with no blocking findings. It also runs scheduled scans for exposed secrets, vulnerable code, misconfigurations and dependencies with known CVEs, and it states SOC 2 Type II compliance.

Autonomous Operations Factory for DevOps, SRE and platform engineering

StackGen offers an 'Autonomous Operations Factory' in which AI agents carry infrastructure work from request to production, with human approval on key actions. Its Aiden agents cover SRE (alert triage, root cause, remediation), InfraOps (Terraform/OpenTofu generation, policy checks, drift remediation, MCP integration with coding tools), Observability and DevOps request triage. Shipyard provides ephemeral environments for developers and their agents.

Trust and infrastructure for open-source ecosystems — now Mecatl.

Stacklok, founded by Kubernetes/Sigstore creator Craig McLuckie, builds trust tooling for open source. In 2026 it shipped Mecatl — an open-source, cloud-native agent harness that architecturally separates the agent loop from code execution, built to run identically on a laptop or Kubernetes.

~$28M (Series A)

4 pros · 4 cons

Open-source agent that lives on your machines 24/7 and keeps your apps running

Stakpak is an open-source DevOps agent written in Rust that runs on your own machines around the clock to keep apps running, contacting a human only when needed. It is positioned as offering the upside of a PaaS without lock-in. Its website states that Stakpak is joining Vercel.

The browser API for AI agents.

Steel Dev provides a batteries-included open-source browser (steel-browser) plus a cloud Browser API for agents and apps — isolated Chrome sessions, stealth, residential proxies, CAPTCHA handling and persistent profiles. The open-source core in the hosted browser-infrastructure lane.

Agentic AI Security for Every Agent

Straiker gives security teams visibility and control across AI agents in the enterprise. Discover AI maps agents, MCP servers and agentic workflows; Ascend AI adversarially tests agents for prompt injection and tool misuse; Defend AI provides runtime detection and blocking of threats to agents.

Security workers and guardrails for code, agents and infrastructure.

Superagent, a YC W24 company based in San Francisco, builds what it calls an autonomous security factory. Its products include PR Security, which checks code, contributor, workflow and supply-chain risk before merge; scheduled dependency updates; Red Team adversarial testing; Context Guardrails, which score what agents consume before they trust it; and Runtime Guardrails, which monitor or block risky agent actions where they run. Findings become reviewed remediation PRs, and it can hand work to coding agents such as Cursor and Devin. Its open-source repo (MIT) targets prompt injection, data leaks and harmful outputs.

Y Combinator (W24)

Memory and context engine for AI, offered as an API, plugins and a local binary.

Supermemory builds a memory and context layer for AI applications and agents. It extracts facts from conversations, handles temporal changes, contradictions and automatic forgetting, maintains user profiles (about 50 ms per call), and combines RAG with memory in hybrid search. It offers connectors for Google Drive, Gmail, Notion, OneDrive and GitHub, and multimodal extractors. It ships plugins and an MCP server that give Claude Code, Cursor, Codex and OpenCode persistent memory, plus a developer API and a self-hostable single binary. TechCrunch reported a $2.6M seed round in October 2025 for founder Dhravya Shah.

$2.6M seed led by Susa Ventures, Browder Capital and SF1.vc (Oct 2025)

4 pros · 4 cons
Acquired

Enterprise AI coding platform and context engine, now part of Tricentis.

Tabnine is an AI coding platform built for secure, context-aware enterprise software development, founded and led by CEO Dror Weiss. Its Enterprise Context Engine maps an organization's code, documentation, tickets and APIs into a continuously refreshed knowledge structure, supporting dependency and impact analysis, governance checks and shared memory for multi-agent workflows, and can run on-premises, in a private cloud or fully air-gapped. On July 30, 2026, testing company Tricentis announced it had acquired Tabnine and will integrate the context engine into its agentic quality engineering platform; tabnine.com now redirects to Tricentis.

Search and extraction API optimized for LLMs and agents.

Tavily gives agents a single API for search, crawling and in-depth extraction, returning clean, cited results — a common tool in research-agent harnesses.

4 pros · 4 cons

Cloud platform that runs any coding agent in shareable sandboxed environments.

Tembo runs AI coding agents (Claude Code, Codex, Cursor, Amp, OpenCode, Pi and others) in isolated cloud environments with source code, terminal, files and a live desktop, turning issues and alerts from Sentry, Linear and Jira into pull requests, including coordinated multi-repo changes. Tembo Previews spin up frontends, APIs and databases behind tunnel links; agents reach databases and team tools via MCP. It offers per-task model routing, scheduled tasks, webhooks, a public API and TypeScript SDK, plus SSO, RBAC, audit trails and self-hosted/VPC deployment.

AI Software Factory for Product Teams

Tempo turns customer feedback, issues and product requirements into reviewed pull requests. AI agents do the work while the team approves key decisions. It is billed as the IDE for the whole team.

Durable execution for long-running agent workflows.

Temporal's durable execution engine is widely adopted to make agent pipelines reliable: state survives crashes, retries are automatic, and workflows run for minutes or months. A key piece of serious harness infrastructure.

$430M total · $300M Series D (Feb 2026) at $5B valuation

3 pros · 4 cons

Launched as "Vercel for filesystem-based agents"; now rebuilds whole departments to run AI-first.

Terminal Use launched on Hacker News in March 2026 as "Vercel for filesystem-based agents", hosting agents that work through files and a shell. It has since pivoted: its site now offers "Whole departments, rebuilt to run AI-first" for construction, logistics, manufacturing and retail. Y Combinator W26.

Agent enablement platform and registry for building, evaluating and distributing agent skills.

Tessl, founded by Guy Podjarny, describes itself as the management layer between developers and their coding agents (Claude Code, Cursor, GitHub Copilot, Gemini and others). The Tessl Platform acts as a control plane for agent context, tackling duplicate skills, tracking skill quality and usage with before-and-after evals, and scanning skills for security. The Tessl Registry indexes skills across an organisation, and Loops run skills on a schedule for tasks like code review and docs maintenance. In September 2025 it launched the Tessl Framework and Spec Registry for spec-driven development.

Open-source AI agents that work for you.

TinyHumans builds a family of open-source agents: OpenHuman, a local-first Rust harness with private lifelong memory; OpenCompany, an agent 'hive mind' that thinks as one team; and Medulla, a terminal that runs Claude Code, Codex and OpenCode side by side.

An AI security and control platform for devices, with an open-source MCP scanner.

Traceforce launched on Hacker News in July 2026 to give companies visibility over the AI apps on their devices and the MCP connections they make. Its site now reads "AI Security & Control Platform for Devices" and offers an open-source MCP scanner. Y Combinator S26.

LLM reliability platform that turns evals and monitors into a continuous feedback loop

Traceloop is an LLM reliability platform for monitoring, evaluating and improving LLM applications. It is built on OpenTelemetry and ships with its open-source OpenLLMetry SDK. It also offers Hub, an OpenTelemetry-based gateway. The site announces that Traceloop is joining ServiceNow.

ByteDance's agentic AI IDE, with the open-source Trae Agent.

TRAE is an AI development environment, available as a desktop download and as TRAE Web, that its site says covers the full development lifecycle from ideation and design through coding and testing. ByteDance publishes the related open-source Trae Agent (bytedance/trae-agent, MIT, Python, about 12k GitHub stars), an LLM-based agent for general-purpose software engineering tasks whose repository lists trae.ai as its homepage. Third-party coverage describes an autonomous SOLO mode that plans work and calls tools; this was not confirmed on TRAE's own pages.

AI SRE for complex systems

Traversal is an AI SRE platform that autonomously diagnoses and fixes production incidents and triages alerts continuously. It uses a causal 'Production World Model' to separate root causes from correlations. It deploys read-only by default, supports bring-your-own cloud and model deployments, and is available via Web, Slack, Teams, MCP and API.

Nerve Center for Agentic Coding

Traycer is a free, open-source desktop app for Mac, Windows and Linux. It runs Claude Code, Codex, Cursor and OpenCode side by side, each in its own worktree. Agents share specs and tickets and can message each other to ask questions, request reviews and hand off work. Users bring their own agent subscriptions.

Open-source durable runtime for building and deploying AI agents and workflows in TypeScript.

Trigger.dev is an open-source (Apache-2.0) platform for long-running AI agents and workflows in TypeScript. Tasks run without timeouts and get durability via checkpointing, retries, queues, concurrency controls, idempotency and versioned deploys, plus human-in-the-loop waitpoints, Realtime streaming of LLM output to frontends, cron schedules and per-run tracing and logs. Its chat agents are pitched as surviving refreshes, redeploys and crashes. It can be used as a managed cloud with usage-based pricing or self-hosted. In December 2025 the company announced a $16M Series A led by Standard Capital with participation from Y Combinator.

$16M Series A led by Standard Capital (Dec 2025)

4 pros · 4 cons

TrueForge: open-source agent harness runtime layer.

TrueFoundry, which sells an enterprise AI gateway and MCP gateway, maintains TrueForge, an open-source (MIT) agent harness that runs the agent execution loop: model calls, MCP tools (with header auth or OAuth), git-backed SKILL.md skills, sandboxed code execution (Daytona today), tool approvals, context management and session state. TrueForge exposes this through a chat UI, an HTTP API with a TypeScript SDK and an embeddable UI SDK, and adds context engineering such as subagents, deferred tool loading, Code Mode and compaction. It runs locally (SQLite) or hosted (Postgres + Redis).

Build locally, delegate to the cloud: runs Claude Code and Codex in cloud sandboxes and returns pull requests.

Twill runs coding agents such as Claude Code and Codex in cloud sandboxes, so a developer can hand off a task from their machine and get a pull request back ("Build locally. Delegate to the cloud."). It is a Y Combinator S25 company and launched on Hacker News in April 2026.

The context layer for agentic software development.

Unblocked merges code, conversations, issues, docs, product data and production systems into one knowledge graph and resolves conflicts between sources using recency, authority and proximity, so agents know what to trust. Its Organizational Memory mines team conventions from PR review feedback and history, results are permission-scoped with source links, and it is accessed via MCP (Cursor, Claude Code, Copilot), CLI, Slack, web and API. The Vancouver company, led by founder and CEO Dennis Pilarinos, raised a $20M USD Series A led by B Capital in May 2025.

$20M USD Series A led by B Capital (May 2025)

3 pros · 4 cons

Vercel's AI agent for building full-stack apps

v0 is an AI agent that turns plain-language descriptions, mockups or screenshots into production-ready code, including full-stack apps and agents. It previews the result in real time, detects and fixes its own errors, uses web search and external tools, and can deploy to Vercel or open a pull request.

On-demand isolated Linux microVMs for running untrusted or AI-generated code

Vercel Sandbox runs arbitrary code in on-demand Firecracker microVMs, each with its own kernel, controlled through an SDK or CLI. Claude, Codex and OpenCode come preinstalled in the base image, so coding agents can run inside a sandbox. It has egress policies with credential injection on outbound requests, snapshots that can start parallel instances, sessions up to 24 hours on paid plans, and persistent Vercel Drives.

Agentic Coding with Multiple Parallel Agents

Verdent AI is an agentic coding platform that runs work from a natural-language description and coordinates parallel agents. It ships as a Mac app and as VS Code and JetBrains extensions. The site positions it for building complete products, including login, data storage, payments and admin tools.

Analytics for AI agents: an SDK that shows product teams how their agents are used.

Voker makes an analytics SDK for teams that ship AI agents, so they can see how agents are used and where conversations fail. It launched on Hacker News in May 2026. Y Combinator S24.

Open-source TypeScript agent engineering platform.

VoltAgent ships a batteries-included open-source TS agent framework — agents, workflows, observability UI and evals — targeted at product teams that want the JS ecosystem's answer to LangGraph.

4 pros · 4 cons

Agentic terminal and open infrastructure for cloud software factories.

Warp, founded in June 2020 by former Google principal engineer Zach Lloyd, began as a Rust-based terminal and now positions itself as open infrastructure for cloud software factories. Teams define a factory in code (factory.yaml plus agent files); agents triggered by issues, chat or schedules take work from triage to a mergeable PR, on Warp's cloud or self-hosted in a VPC, and can be steered from web, mobile, terminal or IDE. It supports any MCP-capable agent, including Claude Code, Codex and Cursor, and bring-your-own model or harness. In April 2026 Warp open-sourced its terminal and its Oz cloud-agent orchestration platform.

$6M seed (GV), $17M Series A (Dylan Field), $50M Series B led by Sequoia (Jun 2023)

3 pros · 5 cons

Observability and continuous improvement for production agents

W&B Weave is Weights & Biases' observability and evaluation product for production AI agents. Its traces are organized around sessions, turns, steps, tools and sub-agents. It adds behavior monitoring with alerts, an evaluation API, guardrail scorers and leaderboards. Coding agents such as Claude Code can connect through W&B skills and an MCP server to read production data and run evaluations.

Spec generation plus autonomous Ralph loops on top of Claude Code or Codex.

Wiggum CLI is a command-line orchestrator for autonomous coding loops, also called Ralph loops. It scans a codebase's stack, generates feature specs through AI-led interviews, and runs plan, implement, test, verify and PR cycles using Claude Code, Codex CLI or another CLI coding agent. An agent mode reads the GitHub backlog, prioritizes issues by label and dependency order, resumes partial branches, reviews diffs against the spec and auto-merges when checks pass. It can run headless in CI. It installs via npm, and the TypeScript source is licensed MIT with the Commons Clause.

A natural-language IDE for building agents.

Wordware (YC) lets domain experts and engineers co-build agents in a document-like natural-language programming environment — 'Notion for prompt-to-agent development'.

$30M seed (2025)

3 pros · 3 cons

Reach the frontier of enterprise growth

WRITER is an enterprise AI platform whose core product, WRITER Agent, executes multi-step work across business systems. It provides Playbooks (natural-language multi-step workflows), reusable Skills, Connectors for acting in business apps, AI Studio for monitoring and governing agents, and an Enterprise Brain context layer, running on its own Palmyra models.

Hosted MCP connecting AI clients to 9,000+ apps

Zapier MCP connects Claude, ChatGPT, Cursor or any AI client to 9,000+ apps. The agent finds the right action and runs it, with auth and permissions managed by Zapier. The GitHub repo is the official plugin distribution for the hosted Zapier MCP server.

Zed

High-performance, multiplayer code editor; open source, AI optional

Zed is a high-performance, multiplayer code editor from the creators of Atom and Tree-sitter. Its site promotes Delta, a multiplayer environment for coding with agents powered by DeltaDB. It also promotes the Agent Client Protocol, which aims to 'connect any client to any agent'. Mistral's Vibe page lists Zed as a host for its coding agent via ACP.

AI orchestration platform with spec-driven coding workflows and IDE agents.

Zencoder (For Good AI Inc.) offers Zenflow Code, pre-built workflows where agents draft specs, implement changes in isolated environments and run verification before review, with parallel execution, scheduled automation and cross-agent review; Zenflow Work, goal-driven automations over Jira, Slack, Notion and other tools; and IDE agents for VS Code and JetBrains. It describes multi-model workflows (plan with Claude, build with Gemini, review with Codex), multi-repo indexing, MCP support, BYOK and SSO.

AI agent security and governance

Zenity is an AI security platform that lets organizations decide what their AI agents can and cannot do. It evaluates every agent action in real time and steers it.

Zep

Long-term memory for agents, built on temporal knowledge graphs.

Zep maintains durable, time-aware agent memory via its open-source Graphiti engine — facts, relationships and their history — so conversations and tasks can span sessions.

4 pros · 4 cons
Shut down

Hosted OpenClaw: a personal OpenClaw agent running on its own cloud VM.

Klaus hosted OpenClaw for people who did not want to run it themselves, giving each user an OpenClaw agent on a dedicated virtual machine. It was one of the first OpenClaw hosting services, and one of the first in that wave to close, as the State of Harness October 2026 report noted.

Shut down

An AI DevOps engineer that fixed CI failures, triaged incidents and reviewed code ("Platform engineering on autopilot").

Mendral automated the DevOps work that follows a pull request: diagnosing CI failures and fixing flaky tests with fix PRs, reviewing dependency updates, triaging incidents and running scheduled automations, with a shared team memory. It was founded by Sam Alba and Andrea Luzzardi and backed by Y Combinator.

Shut down

An agent environment from the lakeFS team: a versioned filesystem, credentials and tools for AI agents.

tilde.run, from the team behind lakeFS, was built on the idea that "AI agents are only as useful as the environment they can work in": a sandbox with a versioned filesystem, credentials, tools and execution for agents. Its lessons went into lakeFS Mount, an agent-native filesystem for production AI workloads.

The Harness Hub newsletter

New harness companies, projects, releases and the monthly State of Harness report, in your inbox.