Show and Tell: ORCH — typed task queue + state machine for AI agent teams (local dev complement to trigger.dev)

[oxgeneral]

Hey trigger.dev community 👋

I've been building a TypeScript CLI runtime for coordinating AI agent teams locally, and found some interesting parallels with what trigger.dev does for background jobs in the cloud.

ORCH addresses the same core problem — reliable async task execution with proper state tracking — but for teams of AI coding agents running on local machines:

Task lifecycle: todo → in_progress → review → done
                              ↓
                           retrying (on failure)

What maps to trigger.dev concepts:

• Task queue → ORCH task store (YAML files, no database)
• Job retries → retrying state with configurable max attempts
• Run logs → append-only JSONL event streams per agent run
• Real-time dashboard → TUI with live event streaming
• Payload/context → shared context store + inter-agent messaging

What's different:

• Agents are long-running processes (Claude Code, Codex, Cursor) not HTTP functions
• No cloud required — all state in .orchestry/ local files
• Mandatory review gate before any task completes — AI outputs need human-or-agent review
• Full TypeScript API: import { Orchestrator } from '@oxgeneral/orch'

Question for trigger.dev users: For AI agent workflows specifically, are you using trigger.dev to coordinate multiple agents in parallel? I'm curious how you handle the "review before done" pattern — do you model that as a separate job, or use something else?

GitHub: https://github.com/oxgeneral/ORCH
npm: npm install -g @oxgeneral/orch