Local agent control plane

Agent Ops

Run coding agents like a control plane, not a chat window

Agent Ops discovers the coding agents already installed on your machine, runs each task in an isolated git worktree, and returns a reviewable diff with tests and logs. You stay the operator.

LiveActive build · migrated from agent-dev· 2026-11-18
AI AgentTypeScriptWorktreeMCP

See how Agent Ops relates to the rest of the Tier 0 family.

Problem

The gap

Coding agents are useful, but their work is scattered across terminals, chats, and local branches. There is no consistent task boundary, no isolated workspace, and no single place to review what changed.

Approach

The control plane

Agent Ops wraps the agents you already use instead of building another runtime. It owns the lifecycle around them: discovery, worktree isolation, run execution, evidence capture, and human acceptance.

What it does

01

Bring your own agent

No new model or harness. Agent Ops schedules the agents already installed on the machine.

02

Task / Run boundary

Each task has its own worktree and run record, so parallel work does not collide.

03

Evidence first

Diff, tests and logs are returned together; a green status is not accepted without them.

04

Human acceptance

Nothing is merged automatically. The operator reviews and decides.

What is real today

Facts taken from the current catalog description and the agent-dev migration plan.

Task model
Task + Run
Isolation
git worktree
Returned evidence
diff, tests, logs
Gate
human acceptance

Boundary

Agent Ops is not a model, a harness, or a chat UI. It is the control surface around external agents; the runtime stays outside.

Where it sits

This product →

replaces · Confirmed

Agent-Dev

Migrated from agent-dev

Evidence (2)
  • src/context/LanguageContext.tsx

    Agent Ops description: migrated from agent-dev.

  • docs/tier0/Tier0-agent-dev并入AgentOps-迁移方案-2026-11-18.md

    Migration plan for folding agent-dev into Agent Ops.

This product →

uses · Proposed

Model Hub

Cost and budget control

Evidence (1)
  • agent-ops/docs/prd/Agent-Ops-多Agent角色与团队-PRD-2026-11-19.md

    PRD M3 future scope: Agent Ops cost and budget control through Model Hub usage data.

This product →

feeds · Proposed

Observability

Run metrics

Evidence (1)
  • agent-ops/docs/prd/Agent-Ops-多Agent角色与团队-PRD-2026-11-19.md

    PRD M3 future scope: run metrics exported to Observability.

See how Agent Ops relates to the rest of the Tier 0 family.

Open the product map