model-configuration（型号配置）

Letta Model Configuration

Patterns for configuring LLM models on Letta agents via SDK/API. Covers model handles, settings, provider-specific configuration, and custom endpoints.

When to Use This Skill

Use this skill when:

• Creating agents with specific model configurations
• Adjusting model settings (temperature, max tokens, context window)
• Configuring provider-specific features (OpenAI reasoning, Anthropic thinking)
• Setting up custom OpenAI-compatible endpoints
• Changing models on existing agents
• Configuring embedding models for self-hosted deployments

Not covered here: Model selection advice (which model to choose) - see agent-development skill's references/model-recommendations.md.

Model Handles

Models use a provider/model-name format:

Basic Model Configuration

Python

from letta_client import Letta

client = Letta(api_key="your-api-key")

agent = client.agents.create(
    model="openai/gpt-4o",
    model_settings={
        "provider_type": "openai",  # Required - must match model provider
        "temperature": 0.7,
        "max_output_tokens": 4096,
    },
    context_window_limit=128000
)

TypeScript

import Letta from "@letta-ai/letta-client";

const client = new Letta({ apiKey: "your-api-key" });

const agent = await client.agents.create({
  model: "openai/gpt-4o",
  model_settings: {
    provider_type: "openai",  // Required - must match model provider
    temperature: 0.7,
    max_output_tokens: 4096,
  },
  context_window_limit: 128000,
});

Common Settings

Context Window Limit

Set at agent level (not inside model_settings):

agent = client.agents.create(
    model="anthropic/claude-sonnet-4-5-20250929",
    context_window_limit=200000  # Use 200K of Claude's context
)

Important:

• Must be <= model's maximum context size
• Default: 32,000 tokens if not specified
• Larger windows increase latency and may reduce reliability
• When context fills up, Letta automatically summarizes older messages

Changing an Agent's Model

Update existing agents with agents.update():

Python

# Change model only
client.agents.update(
    agent_id=agent.id,
    model="anthropic/claude-sonnet-4-5-20250929"
)

# Change model and settings
client.agents.update(
    agent_id=agent.id,
    model="openai/gpt-4o",
    model_settings={
        "provider_type": "openai",
        "temperature": 0.5
    },
    context_window_limit=64000
)

TypeScript

// Change model only
await client.agents.update(agent.id, {
  model: "anthropic/claude-sonnet-4-5-20250929",
});

// Change model and settings
await client.agents.update(agent.id, {
  model: "openai/gpt-4o",
  model_settings: {
    provider_type: "openai",
    temperature: 0.5,
  },
  context_window_limit: 64000,
});

Note: Agents retain memory and tools when changing models.

Provider-Specific Settings

For OpenAI reasoning models and Anthropic extended thinking, see references/provider-settings.md.

Custom Endpoints

For OpenAI-compatible endpoints (vLLM, LM Studio, LocalAI), see references/custom-endpoints.md.

Embedding Models

Required for self-hosted deployments (Letta Cloud handles automatically):

agent = client.agents.create(
    model="openai/gpt-4o",
    embedding="openai/text-embedding-3-small"
)

Common embedding models:

• openai/text-embedding-3-small (recommended)
• openai/text-embedding-3-large
• openai/text-embedding-ada-002

Anti-Hallucination Checklist

Before configuring models, verify:

• Model handle uses correct provider/model-name format
• model_settings includes required provider_type field
• context_window_limit is set at agent level, not in model_settings
• Provider-specific settings use correct nested structure (see references)
• For self-hosted: embedding model is specified
• Temperature is within valid range (0.0-2.0)

Example Scripts

See scripts/ for runnable examples:

• scripts/basic_config.py - Basic model configuration
• scripts/basic_config.ts - TypeScript equivalent
• scripts/change_model.py - Changing models on existing agents
• scripts/provider_specific.py - OpenAI reasoning, Anthropic thinking