ORCH — CLI orchestrator I built to manage multiple AI agents running on local models

[oxgeneral]

What I built

ORCH — a CLI runtime for coordinating teams of AI agents from one terminal.

The problem it solves for LocalAI users

Many LocalAI users run multiple AI processes — code generators, reviewers, shell automators — all hitting local inference. The missing layer: who tracks whether a task is actually done vs just "ran without error"?

ORCH adds a managed lifecycle:

todo → in_progress → review → done
        ↑                ↑
     auto-retry      mandatory gate

Agents are configured with adapters (Claude Code, OpenCode, shell), assigned tasks, and the orchestrator handles restarts and state transitions.

Why this matters for local setups

When you run 5+ agents against local models, silent failures are the norm — processes die, context fills, the agent hangs. ORCH detects stalled processes via PID tracking and moves them back to retrying state automatically.

Details

• TypeScript, Node 20+, MIT
• 1647 tests
• TUI dashboard (Ink/React)
• Works with any local model via shell adapter or OpenCode
• npm install -g @oxgeneral/orch

Repo: https://github.com/oxgeneral/ORCH

Happy to discuss how ORCH could better integrate with LocalAI-backed agent setups.