AO Operator
English | 日本語 | 简体中文 | 繁體中文 | 한국어 | Español | Русский | Français | Deutsch | Português
Midnight. The backlog is growing, and the release is tomorrow. You close your laptop. While you sleep, AO Operator turns your local developer machines into a secure, role-disciplined software factory.
No loose prompts. No context bloat. Every task brief is bound to a strict, verified obligation ledger, executed inside isolated local sandboxes. Wake up to complete, peer-reviewed, and cryptographically verified pull requests.
Safe. Accountable. Done. AO Operator. The local multi-agent software factory. Run it tonight.
Watch The Promo
Video: Watch the 40-second AO Operator promo on YouTube
AO Operator is the AI Orchestration Operation layer: describe the outcome in
natural language, and it drives Codex or Claude Code toward a verified
deliverable. Give it a product request, an SDD, or a task brief. It turns that
intent into scoped roles, cross-platform checks, RunSpecs, status artifacts, and
evidence you can review.
Start here when you want an AI CLI to carry work to done instead of leaving you
with a chat transcript to babysit. AO Operator is built for result-oriented
work: create an app sample from an engineering spec, improve a repo over time,
validate behavior on macOS/Ubuntu/Windows, and keep each role accountable for
evidence before the closer accepts the run.
AO Operator is also the product layer for the wider AO adapter surface:
OpenClaw can submit, schedule, and observe work; Hermes-style queues can drive
worker-saturation runs; AO Runtime keeps provider dispatch, policy, events, and
evidence underneath. The operator gives those plugin and adapter flows a
consistent role contract instead of making every integration invent its own
workflow semantics.
Paste Into Codex Or Claude Code
Do not start by copying shell commands into a terminal. Start in the AI CLI you
already use. Open Codex CLI or Claude Code in a parent directory where a
new checkout can be created, then paste this prompt:
Try AO Operator without spending live provider tokens.
Goal:
- Clone https://github.com/uesugitorachiyo/ao-operator.git if it is not already present.
- Enter the repo.
- Read examples/ingestible-specs/financial-citation-audit-sdd.md.
- Materialize that SDD with the smoke-test profile using the provider-free
ingestion path.
- Do not set OPENAI_API_KEY or ANTHROPIC_API_KEY.
- Stop and explain the blocker if Python 3 or git is missing.
Report back with:
- the workflow outcome requested by the SDD;
- the public wedge AO Operator is proving;
- the role graph AO Operator created;
- the generated RunSpec path;
- the status directory path;
- what the live Codex/Claude execution would do next.
Expected report shape:
requested outcome: financial citation audit workflow
public wedge: citation and compliance review with signed paper trail
profile loaded: smoke-test
role graph: intake -> test-engineer -> evaluator-closer
runspec: run-artifacts/ingest-financial-citation-audit-sdd/ingest-financial-citation-audit-sdd.runspec.yaml
That first run is provider-free. It shows the value before spending live Codex
or Claude Code tokens: AO Operator ingests a high-level product request, chooses
roles, writes a RunSpec, and materializes the evidence paths needed to carry the
work forward. Live execution uses your existing local codex or claude CLI
login, so subscription users can drive structured multi-role work without
managing API keys.
Want the fastest useful trial? Paste the financial-services SDD-style brief and
let AO Operator turn it into a local agent team by asking Codex or Claude Code:
In the AO Operator repo, ingest the financial-services SDD at
examples/ingestible-specs/financial-citation-audit-sdd.md with the smoke-test profile.
Then enter the sibling financial-services profile repo and run:
python3 -m financial_services_profile.cli public-proof --run-id public-sec-citation-proof
Summarize the requested outcome, generated role graph, RunSpec path, status
directory, `public-proof.json`, `public-proof.md`, evidence-pack path, and
verify/replay verdicts.
