deepthinklite

Deep research progress is saved as reusable artifacts by generating questions.md and response.md directly in your local repository. This local-first workflow ensures your investigation map and final answers are durable and version-controlled.

A time-budgeted research workflow enforces configurable minimum and maximum time limits to ensure deep research thoroughness. These guardrails prevent both premature termination and excessive processing time on a single investigation.

Prevent prompt injection during deep research by using a workflow with explicit user permission prompts and untrusted-source handling. It supports raw or summary-only source input modes to actively mitigate data leakage risks.

Yes, deep research works with local-first workflows by keeping all progress and generated artifacts in your local repository. This approach avoids losing chat history and ensures your research outputs remain fully version-controlled.

The best way to structure deep research outputs for version control is a two-artifact model: questions.md for the investigation map and response.md for the final answer. These markdown files are easily diffable within any repository.

Yes, you can use summary-only source inputs for deep research. The workflow supports both raw snippets and summary-only modes, explicitly requiring user permission prompts before processing untrusted sources to ensure safety.