LOCAL-FIRST · MACOS · LINUX · WSL2

Stop babysitting agents.
Run them in parallel.

AOP is the missing orchestration layer between your coding-agent CLI and team-scale delivery. Workflows, worker swimlanes, isolated worktrees, live logs — all on one local dashboard.

MACOS · LINUX · WSL2
$ curl -fsSL https://getaop.com/install.sh | sh

Prebuilt binary (CLI + local-server + dashboard). Requires Git and at least one supported agent CLI on your PATH for live runs.

6 free workers·5 supported agents·12 step types·1 local SQLite

WORKS WITH YOUR FAVORITE AGENT CLI

Install at least one on your PATH before marking tasks Ready.

Think in workers, not in tmux tabs.

Drag tasks onto worker swimlanes. Each row has its own default workflow, repo memberships, and concurrency. The Pool board is your operator console.

Swimlanes

Each worker is a row. Drag a task to assign it.

Three columns

To do · In progress · Done. One status truth.

Real-time

SSE on /api/events. No refresh, ever.

Replace your CLI skill chains with a workflow graph.

Today you run implement → tests → review → debug as a sequence of manual CLI invocations. AOP runs that graph for you, step by step, retries included.

implementrun-testsdebugreviewcleanupdone

aop-default · bundled pipeline · editable in the workflow builder

Run several tickets at once. In separate worktrees.

Each worker runs in its own git worktree under ~/.aop/worktrees/. No merge conflicts, no context bleed, no sibling-agent collisions.

Pool6 workers · 6 tasks
tdd-worker
WORKINGcodex

Add auth middleware

READYcodex

Fix rate limiter

review-worker
WORKINGomp

Review PR #142

DONEomp

Spec cleanup

ci-fix-worker
WORKINGpi

Fix CI on main

READYpi

Retry e2e suite

Live logs. Stop tailing terminals.

dashboard.aop.local · /tasks/:id
Log history
Log history
  • Server-Sent Events

    Same connection, every step, every retry.

  • Step history

    Replay past executions while other workers keep going.

  • One source of truth

    Task docs, runtime events, PR/CI state, all attached to the run.

Annotate the plan before it ships to the agent.

Highlight spec text, batch corrections, let a plan-review agent rewrite the docs. Stop losing feedback in Slack.

Reusable step blocks. One signal, one retry policy, every workflow.

dashboard.aop.local · /workflows
Step blocks
Step blocks

step-block.example.yaml

name: implement
runtime: codex-cli      # or: omp · pi · opencode · agent
prompt:  ./prompts/implement.md
signals:
  done:  "Implementation complete"
  retry: "tests failing"
retries: 2

The bundled methodology. Open the source any time.

aop-default ships TDD, systematic debugging, simplification, and architecture passes as workflow steps. Prompts live in your repo, not in a global skills folder.

implement

Write code against the task spec and plan.

runtime: codex-cli

test

Run the test suite and capture failures.

runtime: pi

review

Review changes for quality and correctness.

runtime: omp

debug

Systematically debug failures with evidence.

runtime: codex-cli

iterate

Refine implementation based on feedback loops.

runtime: opencode

research

Explore the codebase before committing to a plan.

runtime: agent

AOP is not another agent. It orchestrates the ones you already love.

Without AOP

  • Manually run CLI skills for each phase: planning, TDD, implementation, tests, review, debugging, cleanup, CI fixes
  • Keep one agent session alive and move context by hand
  • Re-type the same prompts for every feature or PR fix
  • Run one ticket at a time because every step needs operator attention
  • Guess what is queued, stuck, or ready for handoff

With AOP

  • Define the phase chain once in the workflow builder; workers execute it step by step
  • AOP keeps task docs, workflow state, logs, and git worktrees attached to the run
  • Reuse step blocks with explicit completion signals, retry limits, and runtime overrides
  • Queue tasks across worker swimlanes and run them concurrently in separate worktrees
  • Use the Pool board, task detail logs, status badges, and PR/CI actions as the operator console

“Each worker profile has its own default workflow. Run a review lane, an implement-test-review pipeline, a CI-fix loop. Concurrently.”

— AOP README

Open source. Local-first. Yours.

MIT licensed

Use, modify, self-host. Forever free for the dashboard, CLI, and local-server.

6 workers free

No AOP accounts. No telemetry. Optional license keys raise the cap.

100% local

SQLite at ~/.aop/aop.sqlite. Your code, your worktrees, your machine.

PlanPriceActive workers
Free$06 workers
Pro$5/mo12 workers
Team$10/moUnlimited

Paid plans use license keys only — no AOP account. Licensing details

One curl. One agent CLI. Your factory floor.

MACOS · LINUX · WSL2
$ curl -fsSL https://getaop.com/install.sh | sh

Prebuilt binary (CLI + local-server + dashboard). Requires Git and at least one supported agent CLI on your PATH for live runs.