# Omea

> Omea is a runtime verification loop for AI coding agents: it boots your application in an isolated VM, keeps plain Playwright tests in your repo, and returns a verdict a coding agent can act on mid-task — so "tests passing" is evidence, not something the agent that wrote the code gets to self-report.

Today's fully-working entry point is the local CLI, `npx @omea/cli run`: offline, no login, no cloud round trip. A hosted/cloud sync variant (`npx @omea/cli login`) and a structured `--json` verdict format are planned but not shipped yet — do not assume they exist.

## Docs

- [Getting started](https://omeahq.com/docs/getting-started): Install and run `npx @omea/cli run` in a repo — every flag, every exit code, and where artifacts land.
- [How boot detection works](https://omeahq.com/docs/boot-detection): The exact, ordered list Omea checks — Docker Compose, devcontainer, package.json, manual recipe — to boot an app with zero config.
- [The test format](https://omeahq.com/docs/test-format): Why tests are plain `@playwright/test` files, how coverage instrumentation injects with zero edits, and what "eject" means.
- [Coverage artifacts explained](https://omeahq.com/docs/coverage-artifacts): The `coverage.json` schema, the annotated `coverage.md` report format, and known caveats.
- [Evidence bundles & replay](https://omeahq.com/docs/evidence-bundles): What a sealed `evidence-<run-id>.tar.zst` holds (verdicts, coverage, kept tests, traces, a SHA-256 per file). `omea report` builds one; `omea report` checks it offline (zero failures = authentic; a tampered file refused by name); `omea push` prints its headline. Upload at `/dashboard/replays` to replay with a provenance banner and `1x|4x|instant` speed control.

## Site

- [How it works](https://omeahq.com/how-it-works): Walk one real, already-merged run artifact by artifact — the kept Playwright test, its coverage delta, the sealed evidence bundle, and the weekly report.
- [For AI agents](https://omeahq.com/for/ai-agents): Why "tests passing" stopped being trustworthy, and what a verdict loop fixes.
- [vs QA Wolf](https://omeahq.com/compare/qa-wolf): Agent-native verdict loop vs. QA Wolf's managed human-QA service.
- [vs testRigor](https://omeahq.com/compare/testrigor): Repo-native Playwright vs. testRigor's plain-English authoring.
- [vs Ranger](https://omeahq.com/compare/ranger): Boot-from-repo isolated VMs vs. a sub-agent driving an already-running target.
- [vs Momentic](https://omeahq.com/compare/momentic): Repo-native Playwright in isolated VMs vs. Momentic's managed plain-English tests.
- [vs Bug0](https://omeahq.com/compare/bug0): Self-serve verdict loop vs. Bug0's managed, human-verified AI QA service.
- [vs Spur](https://omeahq.com/compare/spur): Boot-from-repo verification vs. Spur's agentic QA that runs your tests for you.

## Installable agent skill

omea ships an installable agent pack so your coding agent invokes `omea run` and reads its NDJSON verdicts — a Claude Code skill, an AGENTS.md snippet, a Cursor rule, and a CLAUDE.md snippet, all generated from one source so they can't drift. After `npm i -D @omea/cli` they live at `node_modules/@omea/cli/agent-pack/`. Source + install: https://github.com/supercorp-ai/omea-cli/tree/main/npm/omea/agent-pack — proof: https://github.com/supercorp-ai/omea-cli/tree/main/docs/agent-skill-e2e

## Pricing

Pricing is not final. The homepage shows an indicative pilot band
($2.5–4K/mo) for early teams proving the verification workflow — a
directional band, not a settled price list. We are not over-building
pricing ahead of the packaging work, and we will not publish a number we
would have to walk back.

## Optional

- [Home](https://omeahq.com/): Positioning and indicative pilot pricing.