Skip to content

Choose an install

Azhi Flow has two ways to run. They run the same code.

Team installLocal mode (experimental)
ForA team, or anything you want to keep runningOne person on one machine
PartsPostgreSQL 16 + pgvector, Temporal and the Azhi server in Docker, plus one or more workersOne process: embedded PostgreSQL (PGlite), Temporal’s dev server, the server and a worker
NeedsGit, Node.js 22+, Docker with ComposeGit, Node.js 22+ (Python for script nodes)
Commanddeploy/install.shnpx azhi up
DataDocker volumes~/.azhi (%USERPROFILE%\.azhi on Windows)
PlatformTeam installLocal mode
LinuxServer and workersYes
Windows 10/11Inside WSL2 (Ubuntu)Yes, native, no WSL
macOSServer in Docker, development worker. Tested and working; workers not certified.Not yet tested
  • Git and Node.js 22 or newer (engines in package.json).
  • Docker with the Compose plugin for the team install.
  • For script nodes, on each worker host: uv (Python 3.12) and/or Bun. The worker reports what it found, and the run plan marks anything missing as a blocker.
  • A model credential: an Anthropic or OpenAI API key, or a Copilot / ChatGPT / Claude plan sign-in. See Use a subscription, not API keys.
CPU4 vCPU (x86_64)
Memory16 GiB (the three containers use about 420 MiB idle; leave room for load)
Disk20 GiB free (images take about 1.7 GiB)
OSLinux 6.x with Docker Engine and the Compose plugin
NetworkOutbound HTTPS to Docker Hub (or a mirror) and the npm registry

Measured on 2026-10-04: server image build 39 s, start to all three containers healthy 16 s (fresh volumes, migrations included). Image pulls were not measured; at 50 Mbit/s a cold install is estimated at 2 to 3 minutes.