// the find
run-llama/llama_deploy
Deploy your agentic worfklows to production
LlamaDeploy is a deployment layer for LlamaIndex agent workflows — you describe your workflow in a YAML config, point it at a Python file, and get an API server with session management, streaming, and a CLI. It targets teams already using LlamaIndex who want to go from 'workflow that runs locally' to 'workflow with an HTTP API' without writing the scaffolding themselves.
The YAML-based deployment config is genuinely useful — you can specify services, env vars, and source (local or git) in one file, which makes CI/CD straightforward. The API server includes session management out of the box, so multi-turn agent conversations don't require you to wire up your own state store. E2E tests cover real scenarios (streaming, hot-reload, HITL, env vars from git) rather than just happy-path mocks. The `llamactl` CLI makes inspecting running deployments and submitting tasks scriptable without needing to write HTTP client code.
The repo is deprecated — the README opens with a CAUTION block directing you to `llama-agents` / `workflows-py` instead, so adopting this is building on something the maintainers have already abandoned. It's tightly coupled to LlamaIndex workflows; if you want to deploy an agent built with anything else, you're out of luck. No mention of autoscaling, resource limits, or multi-replica deployments — it's essentially a single-process API server with no story for horizontal scale. The Python SDK client is thin enough that you'd outgrow it quickly and end up calling the raw HTTP API anyway.