Open sourcePython · ≈26k stars · Apache-2.0
Letta
Stateful agents with memory, beyond the context window.
Letta (the company behind MemGPT) builds an agent server where agents have persistent memory, self-editing state and tool use — the reference open-source take on memory as a first-class part of the harness.
What the community says
User reports from Hacker News, Reddit, GitHub and reviews, as of Oct 29, 2026. Not our verdict.
Pros people report
- MemGPT/Letta-style memory makes long-context documentation work manageable.Hacker News
- Trusted by some managers as a base to build on, over newer personal-agent hype.Hacker News
- ADE lets you inspect and edit an agent's memory while it runs; Apache-2.0, self-hostable (review, not hands-on).promptquorum.com
Cons people report
- Memory architectures like Letta's 'simply move the pain to retrieval'.Hacker News
- LLM-based memory extraction gets costly at scale ($1k-3k per 100K memories/month, per a competitor's estimate).Hacker News
- Self-hosting and provider bugs: Azure models broken in a release, summariser not triggered for Google models, ADE failing behind a prefix.GitHub issue letta-ai/letta#2582GitHub issue letta-ai/letta#2541GitHub issue letta-ai/letta#2697
- Some find early MemGPT source more instructive than the later, more abstracted framework.Reddit (archived snapshot)
- Overkill for stateless apps; not built for multi-agent coordination.promptquorum.com
- Best for
- Long-lived agents that must learn and remember across sessions.
- Watch out for
- Genuine third-party feedback is thin; much praise comes from Letta staff. Test retrieval quality and memory costs.
Adopting an open-source harness? Read the best-practices guide.
Activity
- Last push
- · 4w ago
- Latest release
- 0.16.8
· 6mo ago - Last 90 days
- 0 releases
0.16.8v0.16.8
· 6mo agoWhat's Changed fix: workflows update by @cpacker in https://github.com/letta-ai/letta/pull/3292 fix(security): use JSON instead of pickle for sandbox-server tool result transport by @kianjones9 in https://github.com/letta-ai/letta/pull/3343 Full Changelog:…
0.16.7v0.16.7
· 7mo agoLetta Server 0.16.7 Release Notes 173 commits since 0.16.6 Released March 31, 2026 Highlights Self-hosted users: this is a big upgrade. The default global context window is raised from 32k to 128k, the context window reset bug (LET-7991) is fixed, and compaction has been…
0.16.6v0.16.6
· 8mo agoHighlights Expanded Conversations API support for default conversation / agent-direct mode. New conversations now initialize with a compiled system message at creation time. Fixed model_settings.max_output_tokens default behavior so it does not silently override existing…
0.16.5v0.16.5
· 9mo agoWhat's Changed chore: bump version 0.16.5 by @carenthomas in https://github.com/letta-ai/letta/pull/3202 Full Changelog: https://github.com/letta-ai/letta/compare/0.16.4...0.16.5
0.16.4v0.16.4
· 9mo agoWhat's Changed fix: update gh templates by @cpacker in https://github.com/letta-ai/letta/pull/3155 chore: release 0.16.3 by @sarahwooders in https://github.com/letta-ai/letta/pull/3158 chore: bump v0.16.4 by @carenthomas in https://github.com/letta-ai/letta/pull/3168 Full…
Checked against GitHub on Nov 21, 2026. Release notes are shortened to plain text.
The team behind it
Berkeley, CA · Est. 2024
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)
In research (8)
- Jetriever: Shifting Agent Memory Retrieval from Similarity to UtilityVerdict: Incremental
Nov 16, 2026 · MemGPT, Letta's predecessor, is a reproduced MAB agentic-memory baseline (28.3 overall, Table 7).
- LEDGER: Auditable Agent Memory Under ChangeVerdict: Useful
Nov 16, 2026 · Descends from MemGPT, cited as prior memory in which the model edits its own memory.
Nov 16, 2026 · MemGPT (Packer et al., 2023), from which Letta grew, is cited as a representative experience-storing memory design.
Oct 27, 2026 · Agent framework built around memory compression and summarisation, the step this paper shows can distort group decisions.
- ETAMem: Decoupling and Aligning Heterogeneous Memory for Long-Term Conversational AgentsVerdict: Incremental
Oct 27, 2026 · The MemGPT/Letta paging memory is cited as a predecessor.
- Stratum: Repository Memory for Coding Agents Beyond the Context WindowVerdict: Incremental
Oct 26, 2026 · Letta (MemGPT) is cited as the closest system in spirit, paging memory without code awareness.
Showing the 6 newest of 8 papers.
Neighbours on the workbench
TypeScript · ≈395k stars · MIT
openclaw/openclaw
The viral open-source personal AI assistant that really does things.
Maintained by OpenClaw Foundation
v2026.11.6releasedJavaScript · ≈308k stars · MIT
affaan-m/ECC
Agent harness performance optimization system: skills, instincts, memory and security for coding agents.
Maintained by Affaan Mustafa
TypeScript · ≈285k stars · MIT
deepseek-ai/deepseek-harness
Everything is a Plugin — DeepSeek's open agent harness.
Maintained by DeepSeek
dsh-v0.2.2-rc.5 prereleasedConnections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
