Guide
AI features
Where the model actually does work in the product, beyond a single 'generate' button.
Models are routed through OpenRouter (Claude, GPT, Gemini, and Llama variants are all selectable, plus a free tier) so a workspace can pick what fits its plan; a few latency-sensitive or vision tasks call Claude directly. Below is what each feature does, grouped by where you'd meet it.
Test generation
From Generate: turns an uploaded PRD/spec, or a pulled ticket from Jira, Zoho Sprints, Trello, ClickUp, or GitHub, into a full suite of structured test cases.
The ticket-readiness gate
A ticket gets scored 0–100 across six dimensions (description clarity, acceptance criteria, scope/endpoints, test data availability, auth requirements, and blocking unknowns) and returns a ready / partial / not_ready verdict. In auto mode, a not_ready verdict transitions the Jira ticket back to To Do and posts an AI-drafted comment naming the specific gaps. This is what stops an under-specified ticket from ever reaching a tester's queue.
The API Runner's ticket-to-test pipeline
The most AI-dense surface in the product: a full chain, each stage callable on its own.
| Stage | What it does |
|---|---|
| QA brief extraction | Turns a ticket or scraped API doc into a structured brief: feature, acceptance criteria, endpoints, sample bodies. |
| Endpoint selection | From a connected Postman collection, picks only the endpoints relevant to the brief. |
| Payload / scenario generation | Builds request payloads and edge-case scenarios from the brief and selected endpoints. |
| AC validation | After execution, checks actual API responses against the ticket's acceptance criteria. |
| Defect drafting | Drafts a Jira defect comment, optionally folding in a Figma field-mismatch comparison. |
Alongside the pipeline: an AI chat co-pilot that can see your endpoints, responses, and brief; a one-click response analysis; and Quick Fix, which diagnoses a failed request (bad body, bad headers, missing auth) and proposes the exact fix.
Self-healing selectors
When a Playwright step's selector fails during a run, the failed step plus a live DOM snapshot of interactive elements goes to a model that returns only a corrected selector, or confirms the original was right and the element is genuinely gone, which means a real bug, not a maintenance chore.
Recording & the browser agent
While recording a flow, AI suggests a robust CSS/XPath selector for each captured step. A separate autonomous browser agent can independently drive a Playwright browser toward a goal against a target URL, with hard caps on steps and token spend so a session can't run away in cost. A related but distinct engine drives existing structured test cases end to end for agentic runs, with TOTP/2FA and BrowserStack device support.
Design and field verification
Given a Figma link with a node ID, TestCatalyst can compare a rendered frame against a live-page screenshot, or extract field-like layer names and diff them against observed API field names: match, naming_mismatch, or missing_in_api. If you only have a screenshot rather than a live Figma link, the same field-extraction works from an image.
Report generation
After a run finishes, an HTML report generates asynchronously in the background. The model tier used depends on plan. See Plans & billing.
AI credits