Letta Code
github.com/letta-ai/letta-code- Category
- AI Agents
- Rank
- No. 317Tools index
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Coding
- Interfaces
- CLI · API
- Builder
- @letta-ai
- GitHub
- 3.4k stars
- Latest release
- v0.33.0
- Date
About
A memory-first coding agent that persists across sessions and learns over time. Unlike traditional session-based coding assistants, it works with a long-lived agent that remembers your preferences, codebase, and past conversations while supporting multiple AI models.
What it does
Letta Code is an agent harness for interactive terminal work and proactive scheduled tasks. Agents can rewrite their context, learn skills, invoke subagents, run hooks, search conversations, and expose the same identity through desktop, browser, mobile, and messaging clients.
Why it's ranked here
The scope is unusually broad and technically concrete. Git-tracked context, programmable skills, permissions, schedules, remote execution, messaging channels, and MCP connections form a coherent agent platform. The main reservation is operational: several cross-machine capabilities depend on signing into Letta, while local backend support is marked experimental.
What's good
Context changes are tracked with Git and can sync to a custom repository. Headless operation supports structured output, making automation practical. Model switching, async or sync subagents, hooks, schedules, searchable conversations, and project-scoped skills cover serious workflows. MCP support spans local processes, HTTP, and server-sent events, including OAuth.
Tradeoffs
The default permission mode is unrestricted, so cautious teams should change it before routine use. Remote environments and managed secrets require a Letta sign-in. The package requires a recent Node runtime, uses Bun for development, and labels its local backend experimental. Persistent, self-editing context also creates governance work that session-only tools avoid.
How to use it well
It best suits engineers who want one agent across terminal sessions, automation runs, machines, and chat channels. Start with a project-scoped skill set, choose a restrictive permission policy, inspect memory regularly, and use headless structured output for jobs. It does not replace an account-free, fully local execution stack for remote routing and shared secrets.
Technical notes+
package.json defines the public npm package, the letta binary, ESM packaging, Apache-2.0 licensing, a Node engine floor, and typed exports for schedules, channels, gateway infrastructure, MCP, memory confinement, and agent presets. src/index.ts drives interactive and headless CLI routing, while src/headless.ts implements streaming output, retries, approval recovery, queued input, remote environment routing, and post-turn memory synchronization. src/mcp-client.ts supports stdio, Streamable HTTP, and SSE transports; src/mcp-oauth.ts adds PKCE-style browser authorization with persisted credentials; src/mcp-runtime.ts registers connected MCP tools into the agent runtime. src/mcp-oauth.test.ts exercises cancellation, discovery, dynamic client registration, callback completion, credential persistence, and reconnection.
Observed
- License
- Apache-2.0
- Primary language
- TypeScript
- Packaging
- Public npm package @letta-ai/letta-code with a global letta executable; repository tooling uses Bun.
- Interfaces
- Interactive CLI, headless CLI with structured output, desktop app, browser and mobile chat, messaging channels, and typed library exports.
- MCP support
- Client connections support stdio, Streamable HTTP, and SSE transports, with OAuth available for network transports.
- Platform support
- Desktop application documented for macOS, Windows, and Linux; Nix and community-maintained Arch Linux installation paths are also documented.
- Test structure
- Tests are colocated with source files, including a dedicated MCP OAuth test file.
Read from README.md, package.json, src/index.ts, src/headless.ts, src/constants.ts, src/mcp-oauth.ts, src/schedules.ts, src/mcp-client.ts, src/mcp-runtime.ts, src/gateway-core.ts, src/agent-presets.ts, src/release-notes.ts, src/channels-slack.ts, src/mcp-oauth.test.ts, src/channels-public.ts.
What it can do
Persist coding agent memory across sessions
Previous coding conversations and interactions → Continuous memory state for future sessions
Learn user coding preferences over time
User coding patterns and feedback → Personalized coding assistance based on learned preferences
Maintain codebase context across conversations
Codebase structure and previous discussions → Contextually aware code suggestions and modifications
Generate code using multiple AI models
Natural language coding requests → Code implementations in various programming languages
Learn and improve coding skills over time
Coding tasks and user feedback → Enhanced coding capabilities and techniques
Provide coding assistance with session continuity
Ongoing coding problems and questions → Consistent help that builds on previous interactions
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.