Agents

Fresh

Source: Letta Code SDK

What is an Agent?

An Agent is a persistent entity in Letta Code that maintains memory across all interactions. Unlike ephemeral sessions, agents accumulate knowledge and improve over time.

Core Concepts

Concept	Description
Agent	A persistent entity with memory, model config, and state
Conversation	A message thread within an agent; multiple can exist
Session	A single execution/connection instance
Default Conversation	Auto-created with the agent; accessible via resumeSession(agentId)

Creating Agents via SDK

Basic Agent

import { createAgent } from '@letta-ai/letta-code-sdk';

const session = await createAgent();

Agent with Custom Memory

const session = await createAgent({
  memoryBlocks: [
    { label: 'persona', value: 'You are a senior backend engineer.' },
    { label: 'human', value: 'The user prefers TypeScript and functional patterns.' },
    { label: 'project', value: 'Working on a Next.js e-commerce platform.' }
  ]
});

Agent with Preset Memory

const session = await createAgent({
  memoryBlocks: 'preset'  // Uses default persona and human blocks
});

Agent with Custom System Prompt

const session = await createAgent({
  systemPrompt: 'You are a meticulous code reviewer who catches bugs early.',
  model: 'claude-opus-4-7'
});

Agent with System Prompt Preset

const session = await createAgent({
  systemPrompt: 'letta-claude'  // Available: default, letta-claude, letta-codex, letta-gemini
});

Agent Lifecycle

Configuration Options

Agent Creation Options

Option	Type	Description
memoryBlocks	Array or 'preset'	Initial memory configuration
systemPrompt	string	Custom system prompt or preset name
model	string	Model to use (e.g., 'claude-opus-4-7')
embedding	string	Embedding model for memory search

System Prompt Presets

Preset	Description
default	General-purpose coding assistant
letta-claude	Optimized for Claude models
letta-codex	Optimized for OpenAI Codex
letta-gemini	Optimized for Google Gemini

Minimal model-specific variants are also available.

Verification Checklist

• [ ] Agent created successfully
• [ ] Memory blocks initialized
• [ ] Agent responds to messages
• [ ] Agent persists state across sessions
• [ ] resumeSession() restores prior context

See Also

• Memory System - How agents store and retrieve knowledge
• SDK API Reference - Full function signatures
• Subagents - Delegate work to specialized agents