Hey everyone, Necip here, the founder. I built aiorch because I kept getting PRs that "AI wrote" but nobody had reviewed. Single-agent tools hand you a diff and the review burden stays with you. aiorch makes the agents review each other before you ever see the diff. Three layers: - An in-loop reviewer that critiques each coder and demands revisions until the work passes spec - An independent auditor with no shared context (adversarial, not confirmatory) - Integration tests across merged branches — no PR opens unless everything passes The harder piece to build was cross-brand routing. Claude Code only runs Claude. Codex only runs OpenAI. That gap won't close — it's a commercial conflict, not a feature one. aiorch decomposes your task and routes each subtask to whichever model is best for the job: Claude, GPT-5, Kimi, Codex, or local Ollama. Mix them in a single run with your own keys. Everything runs in your Docker container. Your code never leaves. Zero telemetry, 0% token markup — you pay providers at list price. I dogfooded this hard while building TravelCalls — a 300k+ line autonomous Rust voice-agent stack for telecom. Multi-day refactors finished overnight as reviewed PRs. That pressure shaped the product. 5 minutes from `docker pull` to your first orchestrated session. Two things I'd love to hear from you: - What kind of work do you currently *not* trust AI agents with because of the review burden? - If you've tried Claude Code or Codex and bounced — what was the trigger? Happy to dig into anything technical.

aiorch

Parallel AI agents that ship reviewed pull requests

Parallel AI agents that ship reviewed pull requests