How letta-ai/letta-code works
Letta Code is a stateful agent harness accessed through a terminal, desktop, browser, or messaging channel. A typical interaction enters through the CLI or a channel, reaches a local or Letta Cloud backend, and produces an agent response; agents can also use tools and subagents. Agent memory, identity, prompts, and skills support long-term adaptation, with MemFS providing Git-tracked context. The README documents major capabilities and entry points; sampled source excerpts are partial, so uncertain internal wiring is omitted.
Main entry points: context.ts Agent Runtime · backend.ts Backend API · App.tsx CLI · service.ts Messaging Channels
Components 21
User Interfaces
- CLI
src/cli/App.tsx - Messaging Channels
src/channels/service.ts - Channel Routing
src/channels/registry-routes.ts - Browser & Desktop
Agent Runtime
- Agent Runtime
src/agent/context.ts - Backend API
src/backend/backend.ts - Local Backend
src/backend/local/local-backend.ts
Agent State
- Memory & Skills
src/agent/memory.ts - Git-Tracked Context
src/agent/memory-git.ts - Settings & Secrets
src/settings-manager.ts
Agent Capabilities
- Tool Execution
src/tools/manager.ts - Subagents
src/agent/subagents/manager.ts - Workflows
src/tools/workflow/workflow-engine.ts - Message Search
src/backend/message-search.ts - Schedules
src/backend/api/schedules.ts - Execution Hooks
src/hooks/writer.ts - Mods
src/mods/mod-engine.ts
Services
- Letta Cloud
- Model Provider
- Remote Computers
src/backend/api/environments.ts
Other components
- User
Connections 21
- User → CLI: interacts
- User → Messaging Channels: messages
- Browser & Desktop → Letta Cloud: accesses agents
- Messaging Channels → Channel Routing: routes messages
- Channel Routing → Backend API: creates conversations
- CLI → Backend API: requests agent turns
- Backend API → Local Backend: dispatches locally
- Backend API → Letta Cloud: dispatches remotely
- Agent Runtime → Backend API: uses backend
- Agent Runtime → Model Provider: requests inference
- Agent Runtime → Memory & Skills: reads and updates
- Memory & Skills → Git-Tracked Context: tracks context
- Agent Runtime → Tool Execution: executes tools
- Tool Execution → Subagents: launches agents
- Workflows → Subagents: schedules calls
- Agent Runtime → Message Search: searches messages
- Agent Runtime → Schedules: configures schedules
- Agent Runtime → Execution Hooks: runs hooks
- Agent Runtime → Mods: loads extensions
- CLI → Settings & Secrets: reads configuration
- Backend API → Remote Computers: routes to computers