Skip to content

FAQ.md

Latest commit

 

History

History
50 lines (35 loc) · 3.69 KB

FAQ.md

File metadata and controls

50 lines (35 loc) · 3.69 KB
Raw

Frequently Asked Questions

What is RepoMind?

A zero‑dependency, offline‑first developer tool that turns a repository into a searchable architecture map, an interactive dashboard, and a repository‑aware AI assistant. See the README.

Do I need an API key or any external service?

No. The MVP runs entirely on the Python standard library. Code search (a lexical BM25 index) and the assistant's grounded, cited answers work with no key and no services. An OpenRouter key is optional and only upgrades the assistant's answers to LLM‑written prose.

What are the requirements?

Python 3.11+. That's it. git is only needed if you want to analyze a GitHub URL (RepoMind shallow‑clones it to a temp directory).

Which languages does it analyze?

  • Python — parsed with the ast module (most detailed).
  • JavaScript / TypeScript / JSX / TSX — lightweight regex parsing.
  • Java — lightweight regex parsing.

Support is defined in analyzer.py (SUPPORTED_EXTENSIONS).

Which API frameworks are detected?

FastAPI, Flask, Express, Next.js route handlers, and Spring (@…Mapping). Detection is heuristic and defined in analyzer.py.

Is the code search "semantic" / embeddings‑based?

No. In v0.1.0 it is lexical (keyword) search: a pure‑Python BM25 index with identifier‑aware tokenization (splits camelCase/snake_case) and structural boosts (matching a file path or symbol name ranks higher). There are no embeddings and no vector database. This keeps the tool key‑free and offline. A true semantic (embeddings) layer is a planned optional backend behind the same search() contract (integrations/vector_store.py). See the Roadmap.

How does the AI chat avoid hallucinating?

Answers are grounded in retrieved code snippets and always include file + line citations. With no LLM key, the assistant returns an extractive answer built directly from retrieval and structural signals. With a key, the model is instructed to answer only from the provided repository context.

How do multi‑turn conversations work?

/api/chat keeps a rolling, capped history in in‑process memory (a thread‑safe dict — no Redis or database). It resolves follow‑up references like "it", "that function", and "this module" against the previously discussed code, and folds recent turns into both retrieval and the prompt. /api/ask remains stateless. Details in ARCHITECTURE.md.

Where is the analysis stored?

Two JSON files under .repomind/: analysis.json (the knowledge map) and index.json (the search index). This folder is git‑ignored.

The chat can't find code I just added — why?

The dashboard reads the on‑disk analysis and index. Re‑run python -m repomind analyze . (or use the dashboard's Analyze button / POST /api/analyze) to refresh them after code changes.

Why not use Neo4j / Qdrant / Docker / FastAPI?

To keep RepoMind usable the instant you clone it, with nothing to set up. Those tools are represented as optional future boundaries (integrations/) and will only ever be opt‑in, never required by the core.

Is my code sent anywhere?

Only if you configure OpenRouter. In that case, the question plus the relevant retrieved snippets and repository summary are sent to OpenRouter to generate an answer. With no key, nothing leaves your machine.

How do I run the tests?

python -B -m unittest discover -s tests -v

Standard‑library unittest, no network required.

How can I contribute or report a bug?

See CONTRIBUTING.md and use the issue templates. The one hard rule: keep it zero‑dependency and offline‑first.