Letta Provider

The Letta AI SDK provider allows you to use Letta agents with the AI SDK.

Features include:

Persistent and long-term memory
Access to both agent-level and model-level reasoning messages with source attribution
Support for custom agent-configured tools and MCP (Model Context Protocol)
Agent-managed filesystem operations with tool-based file access
Built-in utilities to convert between Letta and AI SDK message formats
Every Letta Send Message API feature, like Long-Running Agent Executions

Supported Models

See the Letta documentation for the complete list of supported models. For models not on the list, you can try configuring Letta to use OpenAI proxy.

Setup

You can install the Letta provider with:

pnpm

npm

yarn

bun

pnpm add @letta-ai/vercel-ai-sdk-provider

Provider Instance

You can import the provider instance lettaCloud or lettaLocal from @letta-ai/vercel-ai-sdk-provider:

// For cloud users
import { lettaCloud } from '@letta-ai/vercel-ai-sdk-provider';

// For self-hosted users
import { lettaLocal } from '@letta-ai/vercel-ai-sdk-provider';

// Create a custom Letta provider
import { createLetta } from '@letta-ai/vercel-ai-sdk-provider';

const letta = createLetta({
  baseUrl: '<your-base-url>',
  token: '<your-access-token>',
});

Basic Usage

Get your API key from the Letta dashboard.

# .env
LETTA_API_KEY=your-letta-api-key

import { lettaCloud } from '@letta-ai/vercel-ai-sdk-provider';
import { generateText } from 'ai';

const result = await generateText({
  model: lettaCloud(), // Model configuration (LLM, temperature, etc.) is managed through your Letta agent
  providerOptions: {
    letta: {
      agent: { id: 'your-agent-id' },
    },
  },
  prompt: 'Write a vegetarian lasagna recipe for 4 people.',
});

console.log(result.text);

Advanced Usage

Provider Options

You can configure Letta-specific settings using the providerOptions.letta parameter:

import { lettaCloud } from '@letta-ai/vercel-ai-sdk-provider';
import { streamText } from 'ai';

const result = streamText({
  model: lettaCloud(),
  providerOptions: {
    letta: {
      agent: {
        id: 'your-agent-id',
        maxSteps: 100,
        includePings: true,
        streamTokens: true,
      },
      timeoutInSeconds: 300,
    },
  },
  prompt: 'Tell me a story about a robot learning to paint',
});

Agent Configuration

Configure agent-specific parameters for message creation. These settings apply to both streaming and non-streaming operations.

Available options:

id (string, required): The ID of your Letta agent
maxSteps (number): Maximum number of agent execution steps
includePings (boolean): Whether to include ping messages in the stream
streamTokens (boolean): Enable token-by-token streaming
background (boolean): Enable background execution for long-running operations
Additional parameters available in the Letta API documentation

Timeout Configuration

Type: number

Default: 1000

Set the maximum wait time (in seconds) for agent responses. This is important for long-running agent operations or when working with complex reasoning chains.

const result = streamText({
  model: lettaCloud(),
  providerOptions: {
    letta: {
      agent: { id: 'your-agent-id' },
      timeoutInSeconds: 300, // Wait up to 5 minutes
    },
  },
  prompt: 'Process this complex task...',
});

Letta agents support custom tools and MCP (Model Context Protocol) servers through provider-executed tools. Once configured on your Letta agent , you can include them in your requests using letta.tool() and Letta handles tool execution.

import { streamText } from 'ai';
import { z } from 'zod';
import { lettaCloud } from '@letta-ai/vercel-ai-sdk-provider';

// Use with streaming
const result = streamText({
  model: lettaCloud(),
  tools: {
    web_search: lettaCloud.tool('web_search'),
    memory_insert: lettaCloud.tool('memory_insert'),
    memory_replace: lettaCloud.tool('memory_replace'),
    core_memory_append: lettaCloud.tool('core_memory_append'),
    my_custom_tool: lettaCloud.tool('my_custom_tool'),
    // Optionally provide description and schema (placeholders only - execution handled by Letta)
    typed_query: lettaCloud.tool('typed_query', {
      description: 'Query with typed parameters',
      inputSchema: z.object({
        query: z.string(),
      }),
    }),
  },
  providerOptions: {
    letta: {
      agent: { id: agentId },
    },
  },
  prompt: 'Tell me a story about a robot learning to paint',
});

Using other Letta Client Functions

The vercel-ai-sdk-provider extends the @letta-ai/letta-client, you can access the operations directly by using lettaCloud.client or lettaLocal.client or your custom generated letta.client

// with Letta Cloud
import { lettaCloud } from '@letta-ai/vercel-ai-sdk-provider';

lettaCloud.client.agents.list();

// with Letta Local
import { lettaLocal } from '@letta-ai/vercel-ai-sdk-provider';

lettaLocal.client.agents.list();

More Information

For more information on the Letta API, please refer to the Letta API documentation.