Open Source CLI + SDK ❤️ Letta

kubectl for Stateful AI Agent Fleets

Declarative YAML. Per-agent memory powered by Letta. Git-native versioning. Canary deployments. One command to deploy your entire fleet.

$ npm install -g lettactl

$ npx playbooks add skill letta-ai/skills --skill fleet-management

Quickstart View Docs

Official community tool — featured in Letta docs →Available on playbooks.com →

Declarative YAML

Define your entire agent fleet in version-controlled YAML. One file, one command, done.

Smart Diff Engine

Only changes what needs changing. Detects drift between your config and the server.

Canary Deployments

Test changes on isolated canary copies before promoting to production.

Multi-Tenancy

Tag-based filtering for B2B and B2B2C. Manage thousands of agents across tenants.

Self-Diagnosis

Agents analyze their own memory health. Detect stale data, redundancy, and missing knowledge automatically.

Programmatic SDK

TypeScript SDK for dynamic agent creation. Build SaaS on top of Letta.

shared_blocks:
  - name: brand_guidelines
    description: "Company voice and style"
    limit: 5000
    from_file: "memory/guidelines.md"
  - name: product_catalog
    description: "Current products and pricing"
    limit: 8000
    from_file: "memory/catalog.md"

shared_folders:
  - path: "knowledge/"
    description: "Company knowledge base"

agents:
  - name: support-agent
    description: "Customer support AI"
    tags: { team: support, tier: production }
    llm_config:
      model: "google_ai/gemini-2.5-pro"
      context_window: 32000
    system_prompt:
      from_file: "prompts/support.md"
    shared_blocks:
      - brand_guidelines
      - product_catalog
    shared_folders:
      - "knowledge/"
    memory_blocks:
      - name: customer_data
        agent_owned: true
        limit: 5000
      - name: ticket_context
        agent_owned: true
        limit: 3000
    tools:
      - archival_memory_search
      - archival_memory_insert
      - send_email
      - search_tickets

  - name: sales-agent
    description: "Outbound sales assistant"
    tags: { team: sales, tier: canary }
    llm_config:
      model: "google_ai/gemini-2.5-pro"
      context_window: 32000
    system_prompt:
      from_file: "prompts/sales.md"
    shared_blocks:
      - brand_guidelines
      - product_catalog
    shared_folders:
      - "knowledge/"
    memory_blocks:
      - name: lead_profiles
        agent_owned: true
        limit: 10000
    tools:
      - archival_memory_search
      - archival_memory_insert
      - search_crm

Write your config

Define agents, memory blocks, tools, and shared resources in declarative YAML.

Deploy with one command

Run lettactl apply and the diff engine figures out exactly what needs to change.

Iterate with confidence

Git-native versioning means you can roll back any change. Canary first if you want.

Ready to manage your fleet?

One YAML file. One command. Your entire agent fleet deployed.