What is Canary?
Canary is an AI-powered QA platform that continuously verifies your web application works the way your users expect. Instead of asking engineers to write and maintain brittle end-to-end test scripts, Canary uses browser-driving agents to explore, test, and monitor your product — and surfaces real Issues when something breaks.
01The core idea
You connect a Property (your web application) and give Canary the credentials it needs to act like a real user. From there, Canary does three things:
- Learns the application. Crawlers and agents map your pages, forms, and flows into an Application Map — the truth layer Canary reasons against.
- Runs Workflows. Each Workflow is a journey through the product (sign up, invite a teammate, check out, etc.). A Run is one execution of that journey; failed assertions become Issues you can triage.
- Heals itself. When a selector drifts or a page changes, the workflow agent recovers, updates its cache, and keeps the suite green without human intervention.
02How tests are authored
Tests live as .canary.md files — one workflow, one journey, one test case, written in human-readable Markdown:
- A Spirit blockquote states the user-visible promise.
- A Journey lists steps as bold lead-ins (
**Navigate to**,**Action — …**,**Assert**,**End**). - Inputs and Scenarios parameterize the same journey across plans, roles, or environments.
canary spec compile turns the Markdown into a runnable Canary Workflow. Agents and humans use the same toolchain — see docs/canary-spec-onepager.md.
03What you get day to day
- Regression Suite - that ensures that your core flows don't break as you continue to ship.
- Issues, not test failures. Each detected problem comes with video, network logs, and a reproducible Run.
- Release QA that gates deploys on the workflows that matter for the changes shipping.
- A Canary CLI that exposes every primitive — tasks, workflows, runs, sandboxes, releases, knobs, flags — so it composes cleanly with scripts and agents.
04Who it's for
Teams that want test coverage to grow with the product rather than slow it down: PMs and designers can describe a journey in plain Markdown. Engineers get a typed, reviewable spec and a self-healing suite that pages them only when something genuinely broke.
05In one sentence
Canary turns "did anything break?" into a question your codebase can answer on every PR, every deploy, and every hour — by combining human-authored Markdown specs with browser-driving agents that learn, test, and heal.