Letta
Stateful agent platform (formerly MemGPT) — now a memory-first coding agent with persistent context.
About
Letta began as MemGPT, the research project on LLM agents with self-editing, OS-style tiered memory (core / recall / archival blocks). In 2026 the project pivoted: the original Python V1 API server is retired to the repo's archive branch, and active development moved to letta-ai/letta-code, a TypeScript agent harness.
Today's product, Letta Code, is a memory-first coding agent: an interactive terminal UI, desktop app, and web app that keep persistent memory, identity, and conversation history across sessions and machines (via optional Letta Cloud), with an Agent SDK for embedding stateful agents in TypeScript applications.
What it can do
- • Persistent, self-editing agent memory across sessions (the MemGPT lineage)
- • Interactive terminal UI plus desktop app (macOS/Windows/Linux) and browser web app
- • Multi-channel agents: Slack, Telegram, Discord, and custom channels
- • MCP tool integration and the open Agent Skills standard (installs Hermes/OpenClaw-style skills)
- • Mods: installable runtime extensions (tools, commands, UI) via `letta install`
- • Subagents, conversation search, scheduling, and headless mode
- • Letta Agent SDK for building stateful agents into TypeScript applications
- • Letta Cloud for syncing agent memory, identity, and conversations across machines
Setup
- Install the current CLI with npm (Node.js required):
npm install -g @letta-ai/letta-code - Launch the interactive terminal UI:
letta, then choose "Continue locally" (or sign in to Letta Cloud) and connect your model providers / API keys or coding-plan auth via the bottom-left menu - Initialize agent memory with
/init, then try:What do you know about me so far? - Extend the agent with mods:
letta install npm:@letta-ai/image-understanding, then activate with/reload - Run non-interactively:
letta -p "commit the changes and push"— headless mode for scripts and CI
Prerequisites
Node.js and npm; an LLM API key (or coding-plan credentials) connected in the app menu. The legacy `pip install letta` package targets the retired V1 server and should not be used for new projects.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 25,116 GitHub stars (log-scaled vs catalog peers); blended with ~666,589 monthly package downloads; last push 32d ago; 12 releases/12mo; 0 open issues. Manual review: IDENTITY PIVOT: active dev moved to letta-ai/letta-code (TS coding agent); MemGPT-era Python server retired to archive branch; new app telemetry unverified. License: Apache-2.0.
★ 25,116 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.3/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · 25,116 GitHub stars (log-scaled vs catalog peers)
- · blended with ~666,589 monthly package downloads
- · last push 32d ago; 12 releases/12mo; 0 open issues
- · manual review: IDENTITY PIVOT: active dev moved to letta-ai/letta-code (TS coding agent); MemGPT-era Python server retired to archive branch; new app telemetry unverified
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.