FAQ
Frequently asked questions about SourcePrep.
Is my code uploaded to the cloud?
No. SourcePrep is local-first software. All indexing, vector storage, and processing happens on your machine. The only time data leaves your machine is if you explicitly configure a cloud LLM (BYOK) or during the one-time license activation check.
Does it work with any editor?
SourcePrep works best with editors that support the Model Context Protocol (MCP), such as Cursor, Windsurf, and Claude Code. There is also a VS Code extension in development. For other editors, you can copy-paste context from the Dashboard or CLI.
Do I need a GPU?
No. The core features (indexing, trace graph, search, and compression) run efficiently on CPU. The built-in embedding model is quantized and optimized for CPU inference. Context compression is built in — structural compression for code runs instantly with no model, and language-aware compression for docs uses a lightweight CPU model.
Why does the AI sometimes ignore the SourcePrep tools and use its own search?
AI agents (like Cascade or Cursor) are trained to find the cheapest, fastest path to an answer. If you ask a textual question (e.g., "audit the copy for mentions of 'cloud'" or "find the word 'Ollama'"), the AI will usually prefer its native grep or file search tools because regex is the right tool for pure text matching.
SourcePrep tools shine for structural and semantic tasks. The AI will naturally reach for SourcePrep when you ask about relationships (e.g., "what breaks if I change this function?"), architecture (e.g., "map out the authentication flow"), or when it needs to compress huge amounts of context. You can always force it to use SourcePrep by saying: "Use the prep tools to..."
How is this different from Cursor's built-in index?
Cursor's index is great, but SourcePrep adds a Structural Code Graph layer (understanding imports, definitions, and calls across the project) which reduces hallucinations. SourcePrep also gives you explicit control over context via Path Weights and Compression, allowing you to fit much more relevant code into the context window than a standard RAG approach.
What data is included in bug reports?
SourcePrep's built-in bug report (dashboard → log console → bug icon) collects only diagnostic metadata: app version, OS info, index stats, pipeline status, and error messages. It never includes source code, file contents, embeddings, LLM prompts, or file paths beyond your project root name. You can preview every field before submitting. If you're offline, the report saves as a local JSON file you can review and email manually. See the full breakdown on our Security & Privacy page.