greploom 0.2.0

Semantic code search with graph-aware context retrieval, built on treeloom

greploom

Semantic code search with graph-aware context retrieval, built on treeloom.

greploom reads a treeloom Code Property Graph (CPG JSON), indexes it for hybrid search (vector embeddings + BM25), and returns structurally-complete context neighborhoods for LLM consumption. Vector search finds the right neighborhood; graph traversal expands it to include callers, callees, imports, and data flow sources.

Installation

pip install greploom           # Core — CLI and search engine
pip install greploom[mcp]      # Adds MCP server (requires fastmcp)

The default embedding model is nomic-embed-text via a local Ollama instance. Any OpenAI-compatible embedding endpoint works via GREPLOOM_EMBEDDING_URL.

Quick Start

# 1. Build a CPG with treeloom
treeloom build src/ -o cpg.json

# 2. Index for search (creates .greploom/index.db)
greploom index cpg.json

# 3. Search
greploom query "where is authentication handled?"

How It Works

Source code
    |
    v
treeloom build --> CPG (JSON)
    |
    v
greploom index --> vector store + BM25 index (SQLite)
    |
    v
greploom query "how is auth handled?" --> context bundle
    |
    v
LLM agent receives focused, graph-aware context

Storage is a single SQLite file using sqlite-vec for vectors and FTS5 for BM25. No server required, no Docker, portable and inspectable.

CLI Reference

greploom index

Build or update the search index from a treeloom CPG JSON file.

greploom index CPG_JSON [OPTIONS]

Arguments:
  CPG_JSON    Path to the treeloom CPG JSON file

Options:
  --db PATH              SQLite database path (default: .greploom/index.db)
  --tier [fast|enhanced] Summary tier (default: enhanced)
  --model TEXT           Embedding model name
  --ollama-url URL       Ollama server URL
  --force                Re-index all nodes, ignoring content hashes

Re-indexing is incremental by default — only nodes whose content has changed are re-embedded. Use --force to rebuild from scratch.

Summary tiers:

• fast — function signatures only; fastest to build
• enhanced — signatures, parameters, callees, decorators, and class methods; better recall

# Index with defaults
greploom index cpg.json

# Use a custom database path and force full re-index
greploom index cpg.json --db /tmp/myproject.db --force

# Point at a non-default Ollama instance
greploom index cpg.json --ollama-url http://gpu-box:11434

greploom query

Search the index and return graph-aware context.

greploom query [QUERY_TEXT] [OPTIONS]

Arguments:
  QUERY_TEXT    Natural language or symbol query (mutually exclusive with --node)

Options:
  --db PATH               SQLite database path (default: .greploom/index.db)
  --cpg PATH              CPG JSON path for graph expansion
  --budget INT            Token budget (default: 8192)
  --top-k INT             Number of search results (default: 5)
  --format [context|json] Output format (default: context)
  --model TEXT            Embedding model name
  --ollama-url URL        Ollama server URL
  --node NODE_ID          CPG node ID for direct lookup (repeatable; bypasses search)

Without --cpg, the query returns ranked search hits with scores and summaries. With --cpg, hits are expanded through the graph and assembled into a context bundle trimmed to the token budget. When --format json is used with text search and --cpg, the output is {"metadata": {...}, "blocks": [...]} including index metadata. The --node JSON output is a bare list of blocks (no metadata envelope, since the index is not consulted).

Use --node to retrieve context for known CPG node IDs directly, bypassing the search step entirely. Requires --cpg.

# Simple search — ranked hits with summaries
greploom query "user authentication"

# Full graph-expanded context, ready for an LLM
greploom query "where is authentication handled?" --cpg cpg.json

# JSON output with index metadata
greploom query "UserService" --cpg cpg.json --format json | jq '.metadata'

# Direct lookup by CPG node ID (no index required)
greploom query --node "function:src/auth.py:10:0:3" --cpg cpg.json

# Narrow token budget for smaller context windows
greploom query "error handling" --cpg cpg.json --budget 4096

greploom serve

Start the MCP server.

greploom serve [OPTIONS]

Options:
  --db PATH                    SQLite database path
  --cpg PATH                   Default CPG JSON path
  --host TEXT                  Host to bind (default: 0.0.0.0)
  --port INT                   Port to listen on (default: 8901)
  --transport [streamable-http|stdio]  MCP transport (default: streamable-http)

# Start the MCP server on default port 8901
greploom serve --db .greploom/index.db --cpg cpg.json

# stdio transport for direct agent integration
greploom serve --transport stdio

MCP Server

The MCP server exposes three tools:

search_code — Search code semantically and return graph-aware context.

Parameters: query (required), cpg_path (required), db_path, budget, top_k

get_node_context — Return graph-aware context for specific CPG node IDs, bypassing search.

Parameters: node_ids (required), cpg_path (required), budget

index_code — Build or update the search index from a CPG JSON file.

Parameters: cpg_path (required), db_path, tier

Example MCP server URL for agent configuration: http://localhost:8901/mcp

Configuration

All settings can be provided via environment variables. CLI flags override environment variables for individual commands.

Variable	Default	Description
GREPLOOM_EMBEDDING_URL	http://localhost:11434	Ollama or OpenAI-compatible endpoint
GREPLOOM_EMBEDDING_MODEL	nomic-embed-text	Embedding model name
GREPLOOM_DB_PATH	.greploom/index.db	SQLite database path
GREPLOOM_TOKEN_BUDGET	8192	Default token budget for context assembly
GREPLOOM_SUMMARY_TIER	enhanced	Summary tier (fast or enhanced)

To use an OpenAI-compatible embedding API instead of Ollama:

export GREPLOOM_EMBEDDING_URL=https://api.openai.com/v1
export GREPLOOM_EMBEDDING_MODEL=text-embedding-3-small

Relationship to treeloom

greploom reads treeloom's CPG JSON format but does not import treeloom at runtime. greploom index reads the CPG JSON to build the search index; greploom query reads both the index and the CPG JSON for graph expansion. Any tool that produces treeloom-compatible CPG JSON will work.

greploom's query output includes structural summaries and graph context (callers, callees, parameters), not raw source code. Source text inclusion is a treeloom CPG concern — when treeloom adds source spans to CPG nodes, greploom will surface them automatically.

Changelog

Version 0.2.0

• --node mode for greploom query: retrieve graph context for known CPG node IDs without running a search query.
• Index metadata: embedding model, greploom version, and timestamps are stored in the index and surfaced in JSON output (text search with --cpg --format json wraps results in {"metadata": ..., "blocks": ...}; --node JSON remains a bare list).
• MCP server: added get_node_context tool (direct node ID lookup, parallel to --node in the CLI).

Version 0.1.0

Initial release — full indexing pipeline (summarize, embed, store), hybrid search with RRF, graph expansion for context neighborhoods, token budget management, CLI (index/query/serve), and MCP server with search_code/index_code tools.

LLM Documentation

This project provides llms.txt and llms-full.txt files following the llmstxt.org specification for LLM-friendly documentation.

License

MIT