getaop.com

What It Is

AOP is a local-first orchestration platform for coding Agents, positioned between “coding-agent CLI” tools and team-level delivery workflows. It is not a new Agent. Instead, it brings the CLIs developers already use—such as claude, codex, opencode, Cursor CLI, Pi, and Oh My Pi—into a unified workflow. Planning, implementation, testing, review, debugging, and CI fixes can all be executed step by step by a local orchestrator.

Core Capabilities

At its core are a local dashboard, worker swimlanes, and a workflow graph. Tasks can be dragged into different worker lanes, and each worker can have its own default workflow, repository membership, and concurrency controls. During execution, each task uses an isolated git worktree under ~/.aop/worktrees/, reducing context leakage and merge conflicts. Workflows support steps such as implement, test, review, debug, iterate, and research, and can reuse prompts, completion signals, retry policies, and runtime overrides through step blocks. Logs are displayed in real time via SSE, and previous steps can be replayed. AOP also supports spec annotations, correction rounds, creating GitHub PRs from tasks, and reordering fix-ci workflows based on CI failure context.

Open Source, Self-Hosting, and Pricing

AOP is licensed under MIT, with its source repository publicly available. The documentation clearly states that it can be used, modified, and self-hosted. The CLI, local-server, and dashboard all run locally. The default service address is http://localhost:25150, data is stored in ~/.aop/aop.sqlite, and there are no accounts or telemetry. The free tier includes 4 active workers; Pro costs 5 USD/month and increases the limit to 8; Team costs 10 USD/month and allows unlimited workers. Paid plans only use a license key, obtained through Lemon Squeezy checkout.

Pros and Cons

Its main strength is its clear product boundary: it addresses the pain points of manually chaining multiple Agent CLIs, scattered logs, and hard-to-manage parallel tasks. Its local-first, open-source, no-account design also lowers privacy and deployment barriers. The drawbacks are just as clear: it is currently in alpha, with official warnings about breaking changes and rough edges; only macOS, Linux, and WSL2 are listed; and users must first install Git and at least one supported Agent CLI. Real-world results also depend on those external CLIs, model accounts, and network conditions.

Who It’s For and Access from China

AOP is best suited for individuals and small teams that already use AI coding CLIs in daily development and want to handle multiple tickets in parallel or standardize TDD, review, and CI-fix workflows. For users who only occasionally use chat-based coding assistants, the learning and setup costs may be relatively high. The collected text does not state the situation for access from China. The availability of the official website, GitHub, Lemon Squeezy, and each Agent CLI in mainland China—both in terms of network access and payments—needs to be verified independently. If necessary, local scripts, tmux, Makefile/Taskfile, GitHub Actions, or other open-source Agent platforms may be considered as alternatives.