Skip to content

cgsim_doc_search.md

Latest commit

 

History

History
84 lines (58 loc) · 2.62 KB

cgsim_doc_search.md

File metadata and controls

84 lines (58 loc) · 2.62 KB
Raw

cgsim.doc_search

Package: cgsim Entry point: cgsim.doc_search Module: cgsim.doc_rag Type: Documentation retrieval — vector similarity search

Purpose

cgsim.doc_search searches the CGSim and SimGrid documentation for conceptual questions, how-to guidance, configuration options, and explanations of simulation behaviour. It uses vector similarity search against a pre-built ChromaDB collection.

Use it for questions about how CGSim or SimGrid works, not questions about live simulation state (use a future cgsim.sim_query tool for that).

Typical questions:

  • "How do I write a CGSim plugin?"
  • "What is the SimGrid netzone model?"
  • "How does CGSim calibrate job wall time?"
  • "What methods must a plugin override?"
  • "How does the real-time monitoring dashboard work?"
  • "What output does CGSim write to SQLite?"

Complements cgsim.doc_bm25 — use both together for best coverage (vector search handles semantic similarity; BM25 handles exact-match and enumeration queries).

Inputs

Parameter Type Required Description
query string Yes Natural-language search query.
top_k integer No (default 5) Number of document chunks to return.

Data source

A ChromaDB persistent collection. Configured via environment variables:

Variable Default Description
BAMBOO_CHROMA_PATH ./chroma_db Path to the ChromaDB persistent directory.
BAMBOO_CHROMA_COLLECTION cgsim_docs Collection name to query.

The ChromaDB client is initialised lazily on the first call and cached on the tool instance. If chromadb is not installed, or the configured path does not exist, the tool returns a human-readable error rather than raising.

The cgsim_docs default is intentionally different from the ATLAS (atlas_docs) and ePIC (epic_docs) defaults so all three corpora can coexist in the same ChromaDB directory.

Output

A text block listing the top matching document chunks with their metadata and distance scores. Each chunk includes:

  • Document text (truncated to _SNIPPET_MAX_CHARS = 500 characters).
  • Source file or URL (from metadata).
  • Distance score (lower = more similar).

Dependencies

  • chromadb — ChromaDB client library (pip install -r requirements-rag.txt)
  • Embedding model consistent with the one used during ingestion (default: all-MiniLM-L6-v2, 384-dimensional vectors)

See also

  • cgsim.doc_bm25 — keyword BM25 search over the same corpus
  • docs/rag.md — RAG pipeline architecture, ingestion, and configuration