AI Orchestration
A local control room for software-agent work.
Overview
AI Orchestration is a local Windows prototype for coordinating software work across repositories. It follows tasks from a proposed plan through implementation, checks, review and delivery evidence.
My contribution
I’m developing this personal project with AI coding assistants to explore how agent work can be inspected and recovered. The focus is task state, isolated worktrees, review sessions and explicit approval gates.
Engineering decisions
- Keep durable records. SQLite stores tasks, jobs and evidence. LangGraph checkpoints support workflow recovery.
- Separate roles from execution. Planning, engineering, QA, review and release are logical responsibilities. An enforced execution limit makes the current workflow sequential.
- Tie approval to the work. The controller associates approvals with a source revision and preserves uncertain external operations for inspection. Reviews use fresh sessions.
Evidence and status
This is an implemented local prototype, with no public hosted demo. Deployment support depends on each repository’s configured project contract.
Orca and Chorus are separate tools I use in development. They are distinct from this project.