Get Iranti running in 10 minutes.
You need Node.js 18+ and a PostgreSQL database with pgvector. Everything else is handled by the setup wizard.
If you want to skip the wizard, pass --defaults --db-url <your_postgres_url> to iranti setup.
Install the CLI
Install Iranti globally. You need Node.js 18 or later.
npm install -g iranti
Run setup
The setup wizard walks you through database onboarding, LLM provider keys, and project binding. Bring your own Postgres with pgvector, or let the wizard guide you through a Docker or managed option.
iranti setup
Supports local Postgres, Docker, Supabase, Neon, and any managed Postgres with pgvector.
Start the instance
Start the local MCP server. Keep this running while you work.
iranti run --instance local
Bind your project
From the root of your project, bind it to the running instance. This writes a .env.iranti file with the server URL, API key, and agent identity.
cd /path/to/your/project iranti project init . --instance local --agent-id my_agent
In a multi-agent setup, give each agent its own --agent-id so writes are attributed correctly.
Wire in your AI tool
Run the setup command for the tool you use. Each one configures the MCP connection and protocol instructions automatically.
iranti claude-setup # Claude Code iranti codex-setup # Codex CLI iranti copilot-setup # GitHub Copilot
Verify the connection
Run the doctor command to confirm the database, API key, and provider are all reachable.
iranti doctor
What happens next
Once your AI tool is wired in, it will call iranti_handshake at session start and iranti_attend before each response. Facts get written after edits and tool calls, and the next session starts with the relevant ones already injected.