Legal document

AI Policy

How ThreadMine uses artificial intelligence — and what we never do with your dumps.

AI-powered analysis is one of ThreadMine's core features: it speeds up reading thread dumps, suggests likely root causes and summarizes detected problems in natural language. That power comes with responsibility, and this policy explains exactly what we do — and what we do not do — with your data when we use AI models.

1. Models we use

ThreadMine integrates models from external providers selectable by the workspace admin. The following are currently supported:

Anthropic Claude (Claude family, latest-generation models).
Google Gemini (Gemini family, latest-generation models).

New providers may be added in the future. When that happens, this policy will be updated and admins will be notified.

2. What we send to the AI

When an AI analysis is triggered, we send the model:

The content of the thread dump you submitted (full text or relevant excerpts).
Technical metadata of the dump: timestamp, detected format, hostname extracted during parsing.
Intermediate results from our analysis engine (e.g. the list of problems detected by the deterministic parser).

We do not send: account data (name, email, password), billing data, information from other workspaces, other users' data or any information not essential to generate the requested analysis.

3. Model training — an explicit commitment

Your dumps are never used to train AI models. Neither ours (we do not train our own models) nor those of the subprocessors. Anthropic and Google have a default policy of not using inputs submitted via API for model training — the data is used only to generate the requested response.

If a provider changes this policy in the future, we will stop using it and migrate to an alternative that maintains this commitment.

4. Opt-out: disabling AI

The workspace admin can completely disable AI analysis in the workspace settings. When disabled, none of your data will be sent to LLM providers. The deterministic analysis engine (parser, problem detection, health score) continues to work normally.

You can also disable AI for a single analysis at upload time, if you have permission to do so in the workspace.

5. Transparency of outputs

Every AI-generated output is stored alongside the corresponding analysis. You can:

View the raw output at /analises/{id}.
Identify which model generated each output (metadata attached to every response).
Delete the entire analysis, including the generated outputs, at any time.

6. Limitations — always review

Generative AI models can be wrong. There are known cases of hallucinations (plausible but false statements), partial conclusions or incorrect interpretations of unusual stack traces. For this reason, never make critical operational decisions based solely on AI outputs. Review the raw data, check the locks/threads cited in the original dump and use the AI as support — not as a verdict.

For CRITICAL severity problems, the interface displays an explicit warning recommending manual validation before any mitigation in production.

7. AI usage audit log

Every call to an AI model generates an entry in the workspace audit log: who triggered it, which analysis, which model, estimated cost in tokens. Admins can inspect this log at /configuracoes/auditoria.

8. Costs and quotas

Each plan has a monthly quota of AI analyses. When the quota runs out, new analyses continue to run on the deterministic engine, without AI, until the next cycle. You will never be charged for AI usage above the quota without explicit consent.

9. Contact

For specific questions about AI usage or to report a problematic output: dpo@threadmine.dev. See also our Privacy Policy and our Terms of Service.