2 pointsby diraigent3 hours ago1 comment
  • diraigent3 hours ago
    I built a platform for running AI coding agents (currently Claude Code) through structured, repeatable pipelines — fully self-hosted.

    The core idea: define multi-step playbooks (implement → review → merge), and let an orchestrator spawn agents in isolated git worktrees for each task. Every step is logged, every action auditable.

    Key design decisions: - Agents run in isolated git worktrees, so parallel tasks don't interfere - Configurable git strategies (merge to default, branch-only, etc.) - Agents build up a persistent knowledge base, decision log, and observations across runs - Human review gates at any pipeline step - Your code, your infra, your API keys — nothing leaves your environment

    Stack: Rust/Axum API, Rust orchestrator, Angular dashboard, PostgreSQL. Deploys via Docker Compose.

    The platform has been running my own workflow for a while now — most of it was built through its own pipelines at this point.

    Source and Docker Compose quickstart: github.com/diraigent/diraigent

    Interested in feedback on the platform