Your specs never leave your network.

QAMind uses Qwen 3 Coder 30B, a code-focused large language model from Alibaba Cloud, run locally via Ollama. The model runs on the workspace owner's workstation — a machine with a GPU capable of running a 30B-parameter model.

QAMind's servers communicate with that workstation over a private VPN (Tailscale). The model is never hosted on QAMind's own infrastructure or on a shared cloud GPU farm.

When you generate test cases, the following data is sent from the QAMind server to your local model:

• The requirement document or feature description you pasted
• The browser recording events (clicks, inputs, navigations) from the QA Recorder
• The test generation prompt (a structured instruction, not your data)

This data goes to your workstation, not to a third-party API. The AI model processes it locally and returns the generated test cases. No copy of your input is retained by the model after the request completes.

Not everything is local. The following is stored on QAMind's hosted infrastructure (Supabase / PostgreSQL):

• Your account credentials (email, hashed password or OAuth token)
• Workspace metadata: project names, suite names, team member roles
• Generated test cases — the output of the AI, after you save them
• Test run results and bug reports

The input to the AI — the requirement document you pasted or the recording you imported — is not stored in QAMind's database. It is passed to the local model and discarded after the response is returned.

Running a 30B-parameter model on a single workstation is slower than a cloud inference endpoint. Generation takes 3–8 minutes per request. The workstation queues up to 100 requests and runs one GPU job at a time — GPT chat, test cases, and script generation all share the same line.

For teams where the privacy of requirement documents matters more than generation speed, this is the right tradeoff. For teams with no compliance constraints, a cloud-inference option may be offered in a future release.

QAMind also includes a private AI chat interface at gpt.qamind.ai — a ChatGPT-style interface powered by the same local Qwen model. Use it to ask QA-related questions, analyse PRD documents, or get help with Playwright assertions. Conversations are stored in your workspace and organized by project.

Like test generation, GPT Chat messages are processed by the local model. Your conversation content is not sent to any external AI service.