AI and your data
Which providers run Qualflare's AI features, what each one receives, what is redacted first, the no-training commitment and the switches that turn AI off.
Last updated
Qualflare's AI features send prompts to three model providers, Google Gemini, OpenAI for Quo on workspaces Qualflare has enabled for it, and Cloudflare's Clef model, and none of them trains its own models on your data; you can turn AI off for each project.
Which provider runs which feature
Qualflare's AI features use three model providers. Requests to Gemini, OpenAI and Clef pass through Cloudflare AI Gateway, which keeps the request and the response in its logs.
| Provider | Features | Where it runs |
|---|---|---|
| Google Gemini | Search embeddings, failure clustering, launch analysis, the AI filter's query generation, and Quo by default | United States and other regions where Google operates |
| OpenAI | Quo, on workspaces Qualflare has enabled for it | United States |
| Clef, Cloudflare's model on Workers AI | Typed decisions: how to route a Quo request, whether your reply approves a change Quo proposed, whether an AI filter matches what you asked, and a failure cluster's category, severity, membership and redundancy | Cloudflare's Workers AI network: global, with EU-region routing; core data planes primarily in the United States |
When Quo answers a question about how to use Qualflare, it sends the search query, as written, to Cloudflare AI Search, which searches the public Qualflare documentation and runs a generation model to write the answer.
The model selector in Quo shows which provider runs each model.
What each provider receives
What each AI provider receives depends on the features it runs:
- Google Gemini: the prompts and outputs of its features, which can include personal data that appears in your test results.
- OpenAI: your message as typed, the conversation history, and the agent's tool calls and their results, which carry test case, defect, failure cluster and launch content. It's sent as written, with OpenAI's retention disabled since 28 September 2026.
- Clef: the last part of a Quo conversation, failure messages, test case names and descriptions, and an AI filter request with the filter generated for it, all redacted first.
LangWatch, an AI observability service in the EU, also receives Quo's traces: prompts, completions and tool calls.
Redaction before sending
Qualflare's redaction is a pattern list that replaces secrets and personal identifiers with labels such as [EMAIL] before a request leaves. It runs on search embeddings, failure clustering and the redundancy check, launch analysis, Clef's typed decisions and the Help improve Qualflare AI records. Quo's messages, conversation history and tool results, and the AI filter's query generation, go to Gemini or OpenAI as written: Qualflare caps their length and strips control characters, but doesn't redact them.
The pattern list matches:
- JSON Web Tokens
- AWS access key IDs
- Stripe keys
- GitHub tokens
- Slack tokens
- Cloudflare API tokens and keys
secret=,token=,password=and similar assignments- Email addresses
- UUIDs
Other text, such as names, hostnames and internal URLs, is kept as written.
No training on your data
None of Qualflare's AI providers trains its models on your prompts, outputs, test artifacts or workspace content. Google's enterprise terms for the Gemini API exclude API inputs and outputs from improving Google's models. OpenAI's API terms exclude data sent through the API from training, and Qualflare's Data Processing Addendum with OpenAI has been in force since 24 October 2026. Cloudflare commits that it doesn't use content processed by Workers AI, where Clef runs, to train AI models. Section 6 of the Privacy Notice and section 15 of the DPA give the detail.
Help improve Qualflare AI
The Help improve Qualflare AI setting on the Workspace Settings page is on by default. While it's on, Qualflare keeps a record of AI requests from your workspace: the prompt, the result, and the failure messages and test names involved. Email addresses and common token, key and password formats are removed first. Other text, such as names, hostnames and internal URLs, is kept as written. Anyone with the Editor, Maintainer or Owner workspace role can turn the setting off, and collection stops within about five minutes. Records collected before then can't be withdrawn, because they carry no workspace or project ID.
The switch is described in Help improve Qualflare AI.
Turn AI off for a project
To turn AI off for a project, switch off Enable AI Features in Project Settings > AI Features; it's on for new projects. Off stops failure clustering, new launch analyses, the redundancy check, Quo and the AI filter for that project, as What stops when AI is off lists.
Related
Data privacy and sub-processors · AI in Qualflare · Quo agent · Privacy Notice · GDPR at Qualflare
Data privacy and sub-processors
Where Qualflare stores customer data, which sub-processors handle it and for what, how you hear about a new one before it starts, and how to accept the DPA.
Data retention and deletion
How long Qualflare keeps test results, uploaded screenshots and videos, exports, notifications and audit log entries, and what deleting an account removes.