Action Llama - Action Llama

Open source and MIT Licensed. Use Claude, OpenAI, or other models to:

Automate your dev workflow: agents handle issues, PRs, monitoring
Automate communications: have an agent create a summary of product changes for the last week and tweet it out

You’re progamming with prompts: any process can be automated using an agent.

First Steps

Zero to running agent in minutes

Deploy, scale, optimize, and orchestrate

Understand agents, scheduling, and locks

Every CLI flag, config field, and command

Docker isolation — each run gets its own container with only the credentials it needs
Git-native — define agents in a git repo, add custom ones, share them across teams
BYOM — bring your own model (Anthropic, OpenAI, Groq, Google Gemini, xAI, Mistral, OpenRouter, or any custom provider)
Deploy anywhere — run locally for development or deploy to Cloud Run for production
Webhook + cron — react to GitHub issues, Sentry alerts, Linear tickets, or poll on a schedule
Multi-agent — agents can call other agents with call_agent and collect results with check_call
Resource locks — coordinate parallel instances with automatic deadlock detection
Web dashboard — live agent status and streaming logs in your browser

Agents are infrastructure: they should be versioned, repeatable, and deployable. Agents work together, so they should be bundled together.
Stateless by default: state is easy to add to a stateless system. Dynamic agent context strategies can be added if desired.
Bring your own models: there are many models out there, each appropriate to a different task. You should be able to use any of them.
Minimize the harness and lean on the model: models are improving constantly, that’s why this framework lets you program with prompts. The harness should focus on orchestration only.