Nous Research
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.
Open-source project
In the news
DeepSeek ships Harness Desktop; the 'everything is a plugin' harness passes 240k stars
DeepSeek releases a macOS and Windows Desktop app for DeepSeek Harness, its open-source agent harness where everything — tools, sub-agents, even the loop — is a plugin. The repo passed 240k stars within three months of its August launch, making it the fastest-growing harness in the field's history.
EverMind open-sources Raven — 'the harness of harnesses'
EverMind AI releases Raven, a trusted, persistent, self-evolving multi-agent ecosystem built for recursive self-improvement — agents that modify their own harness under governance. The Show HN thread became a running debate on how much autonomy a harness should hand an agent over itself.
PrimeIntellect's Prime Agent: a self-improving RLM agent
PrimeIntellect open-sources Prime Agent, a recursive-language-model agent whose harness runs sub-agents as function calls inside a persistent IPython REPL. The launch post reports its harness taking Opus 5 from ~30% to 95.5% on ARC-AGI-3 — the sharpest evidence yet that harness engineering can beat model selection.
xAI open-sources Grok Build, its coding agent harness
xAI open-sources Grok Build — a Rust coding agent harness and fullscreen TUI around Grok models — unusually early for a first-party tool. It joins Claude Code, Codex and Gemini CLI as the fourth major first-party harness, now with a fourth open one.
Related startups
Letta
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.
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.
Daytona
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.