# Letta Code SDK

[![npm](https://img.shields.io/npm/v/@letta-ai/letta-code-sdk.svg?style=flat-square)](https://www.npmjs.com/package/@letta-ai/letta-code-sdk) [![Discord](https://img.shields.io/badge/discord-join-blue?style=flat-square&logo=discord)](https://discord.gg/letta)

> [!TIP]
> Check out [**LettaBot**](https://github.com/letta-ai/lettabot) and [**Letta Cowork**](https://github.com/letta-ai/letta-cowork), two open-source apps built on the **Letta Code SDK**.

The SDK interface to [**Letta Code**](https://github.com/letta-ai/letta-code). Build agents with persistent memory that learn over time.

```typescript
import { createSession } from '@letta-ai/letta-code-sdk';

const session = createSession();
await session.send('Find and fix the bug in auth.py');
for await (const msg of session.stream()) {
  if (msg.type === 'assistant') console.log(msg.content);
}
```

## Installation

```bash
npm install @letta-ai/letta-code-sdk
```

## Quick Start

### One-shot prompt

```typescript
import { prompt } from '@letta-ai/letta-code-sdk';

// One-shot (uses default agent - like `letta -p`)
const result = await prompt('What is 2 + 2?');
console.log(result.result);

// One-shot with specific agent
const result2 = await prompt('Run: echo hello', agentId);
```

### Multi-turn session

```typescript
import { createAgent, resumeSession } from '@letta-ai/letta-code-sdk';

// Create an agent with custom memory (has default conversation)
const agentId = await createAgent({
  memory: ['persona'],
  persona: 'You are a helpful coding assistant for TypeScript projects'
});

// Resume the default conversation
await using session = resumeSession(agentId);

await session.send('What is 5 + 3?');
for await (const msg of session.stream()) {
  if (msg.type === 'assistant') console.log(msg.content);
}

await session.send('Multiply that by 2');
for await (const msg of session.stream()) {
  if (msg.type === 'assistant') console.log(msg.content);
}
```

### Persistent memory

Agents persist across sessions and remember context:

```typescript
import { createAgent, resumeSession } from '@letta-ai/letta-code-sdk';

// Create agent and teach it something
const agentId = await createAgent();
const session1 = resumeSession(agentId);
await session1.send('Remember: the secret word is "banana"');
for await (const msg of session1.stream()) { /* ... */ }
session1.close();

// Later... resume the default conversation
await using session2 = resumeSession(agentId);
await session2.send('What is the secret word?');
for await (const msg of session2.stream()) {
  if (msg.type === 'assistant') console.log(msg.content); // "banana"
}
```

## Multi-threaded Conversations

Run multiple concurrent conversations with the same agent. Each conversation has its own message history while sharing the agent's persistent memory.

```typescript
import { createAgent, createSession, resumeSession } from '@letta-ai/letta-code-sdk';

// Create an agent (has default conversation)
const agentId = await createAgent();

// Resume the default conversation
const session1 = resumeSession(agentId);
await session1.send('Hello!');
for await (const msg of session1.stream()) { /* ... */ }
const conversationId = session1.conversationId; // Save this!
session1.close();

// Resume a specific conversation by ID
await using session2 = resumeSession(conversationId);  // auto-detects conv-xxx
await session2.send('Continue our discussion...');
for await (const msg of session2.stream()) { /* ... */ }

// Create a NEW conversation on the same agent
await using session3 = createSession(agentId);
await session3.send('Start a fresh thread...');
// session3.conversationId is different from conversationId

// Start fresh conversation with default agent
await using session4 = createSession();
```

**Key concepts:**
- **Agent** (`agentId`): Persistent entity with memory that survives across sessions
- **Conversation** (`conversationId`): A message thread within an agent
- **Session**: A single execution/connection
- **Default conversation**: Always exists after `createAgent()` - use `resumeSession(agentId)` to access it

Agents remember across conversations (via memory blocks), but each conversation has its own message history.

## Session Configuration

### System Prompt

Choose from built-in presets or provide a custom prompt:

```typescript
// Use a preset
createSession(agentId, {
  systemPrompt: { type: 'preset', preset: 'letta-claude' }
});

// Use a preset with additional instructions
createSession(agentId, {
  systemPrompt: { 
    type: 'preset', 
    preset: 'letta-claude',
    append: 'Always respond in Spanish.'
  }
});

// Use a completely custom prompt
createSession(agentId, {
  systemPrompt: 'You are a helpful Python expert.'
});
```

**Available presets:**
- `default` / `letta-claude` - Full Letta Code prompt (Claude-optimized)
- `letta-codex` - Full Letta Code prompt (Codex-optimized)
- `letta-gemini` - Full Letta Code prompt (Gemini-optimized)
- `claude` - Basic Claude (no skills/memory instructions)
- `codex` - Basic Codex
- `gemini` - Basic Gemini



## API Reference

### Functions

| Function | Description |
|----------|-------------|
| `createAgent(options?)` | Create new agent with custom memory/prompt. No options = blank agent with default memory blocks. Returns `agentId` |
| `createSession(agentId?, options?)` | New conversation on specified agent. No agentId = uses LRU agent (or creates "Memo" if none exists) |
| `resumeSession(id, options?)` | Resume session - pass `agent-xxx` for default conv, `conv-xxx` for specific conv |
| `prompt(message, agentId?)` | One-shot query with default/specified agent (like `letta -p`) |

### Session

| Property/Method | Description |
|-----------------|-------------|
| `send(message)` | Send user message |
| `stream()` | AsyncGenerator yielding messages |
| `close()` | Close the session |
| `agentId` | Agent ID (for resuming later) |
| `sessionId` | Current session ID |
| `conversationId` | Conversation ID (for resuming specific thread) |

### Options

**CreateAgentOptions** (for `createAgent()` - full control):

```typescript
// Create blank agent
await createAgent();

// Create agent with custom memory and system prompt
await createAgent({
  model: 'claude-sonnet-4',
  systemPrompt: 'You are a helpful Python expert.',
  memory: [
    { label: 'persona', value: 'You are a senior Python developer' },
    { label: 'project', value: 'FastAPI backend for a todo app' }
  ]
});
```

**CreateSessionOptions** (for `createSession()` / `resumeSession()` - runtime options):

```typescript
// Start session with permissions
createSession(agentId, {
  permissionMode: 'bypassPermissions',
  allowedTools: ['Bash', 'Glob'],
  cwd: '/path/to/project'
});
```

**Available system prompt presets:**
- `default` / `letta-claude`, `letta-codex`, `letta-gemini` - Full Letta Code prompts
- `claude`, `codex`, `gemini` - Basic (no skills/memory instructions)

### Message Types

```typescript
// Streamed during receive()
interface SDKAssistantMessage {
  type: 'assistant';
  content: string;
  uuid: string;
}

// Final message
interface SDKResultMessage {
  type: 'result';
  success: boolean;
  result?: string;
  error?: string;
  durationMs: number;
  conversationId: string;
}
```

## Examples

See [`examples/`](./examples/) for comprehensive examples including:

- Basic session usage
- Multi-turn conversations
- Session resume with persistent memory
- **Multi-threaded conversations** (createSession, resumeSession)
- System prompt configuration
- Memory block customization
- Tool execution (Bash, Glob, Read, etc.)

Run examples:
```bash
bun examples/v2-examples.ts all

# Run just conversation tests
bun examples/v2-examples.ts conversations
```

## Internals

### CLI Mapping

The SDK spawns the Letta Code CLI as a subprocess. Here's how API calls map to CLI flags:

| Function | CLI Flags | Behavior |
|----------|-----------|----------|
| `createSession()` | `--new` | LRU agent + new conversation |
| `createSession(agentId)` | `--agent X --new` | Specified agent + new conversation |
| `createAgent()` | `--new-agent` | New agent + default conversation |
| `resumeSession(agentId)` | `--agent X --default` | Specified agent + default conversation |
| `resumeSession(convId)` | `--conversation X` | Derived agent + specified conversation |
| `prompt(msg)` | *(none)* | LRU agent + default conversation |
| `prompt(msg, agentId)` | `--agent X --new` | Specified agent + new conversation |

**Key concepts:**
- **LRU agent**: Most recently used agent from `.letta/settings.local.json`, or creates "Memo" if none exists
- **Default conversation**: The agent's primary message history (always exists)
- **New conversation**: Fresh message thread, isolated from other conversations

## License

Apache-2.0