ORCH — orchestration layer for AgentOps-monitored multi-agent pipelines

[oxgeneral]

Hey AgentOps community — sharing a tool that pairs well with AgentOps for multi-agent lifecycle management.

The problem: AgentOps gives you excellent observability into what your agents did — LLM calls, costs, error traces. But the coordination layer — routing work between agents, handling failures, enforcing review gates before task completion — still requires manual wiring. You can see the agents failed in AgentOps, but there's no runtime that automatically retries them or queues the next agent.

ORCH adds that coordination layer on top: a CLI runtime with a validated state machine (todo → in_progress → review → done), auto-retry on agent failure (configurable attempts), inter-agent messaging, and a shared context store. ORCH handles lifecycle; AgentOps handles observability — they're complementary. Each ORCH-managed agent can be instrumented with AgentOps SDK calls.

npm install -g @oxgeneral/orch
orch agent add "CodeWriter" --adapter claude --role "Write TypeScript code"
orch agent add "CodeReviewer" --adapter claude --role "Review and approve code"
orch task add "Add auth middleware" -d "JWT validation" --assignee CodeWriter
orch run --all
# Each run → visible in AgentOps dashboard via SDK instrumentation

State machine ensures no task reaches done without passing through review. Agents in isolated git worktrees. MIT license, 1647 tests.

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

Curious if others here are combining AgentOps observability with an orchestration layer for production workflows?