Choose an Agent and Model
Generation settings live in Settings → AI Model. They control how much reasoning effort the generation agent spends per generation and how long a single generation is allowed to run.
AI runs on TestVibe's metered first-party service — usage is billed in credits and you do not need to provide a provider API key.
The Settings
| Setting | What it controls |
|---|---|
| Model | The AI model TestVibe uses for feature drafting and Playwright code generation. TestVibe currently offers a single first-party model, so there is nothing to pick — the field simply names it. |
| Reasoning effort | Low, Medium, or High. Higher effort produces more thorough tests but uses more credits and takes longer. |
| Generation timeout (minutes) | The maximum time one generation may run before TestVibe stops it and marks the feature failed rather than shipping an unfinished test. Empty uses the 60-minute default; values are clamped to 5–120 minutes. |
New projects default to Medium reasoning effort and the 60-minute generation timeout.
Tests generate with the TestVibe Assistant — reasoning effort maps to the underlying provider's effort setting. Self-hosted installs configured with a custom OpenAI-compatible endpoint see their own model list here instead; see AI models (self-hosted) if that applies to your install.
A generated test only reaches generated status after TestVibe replays the whole assembled spec end-to-end and it passes — this validation always runs and cannot be turned off.
When To Keep Defaults
Keep the defaults when:
- you are doing the quickstart
- your team already configured the project
- generation is producing acceptable output
- you want consistent results across the team
When To Change
| Situation | Try |
|---|---|
| Output quality is weak on complex flows | Raise reasoning effort to High. |
| Generation feels slow or uses too many credits for routine tests | Lower the reasoning effort. |
| A heavy feature is hitting the generation timeout | Raise Generation timeout (minutes), up to 120. Most features finish well under the 60-minute default. |
After Changing
Run a small generation first. Compare the output against the same expectations:
- Does it follow the Gherkin?
- Does it use the correct UI?
- Does it include real assertions?
- Does it generate code your team can review?
If the new settings work well, save them as the project default.