Configuration

Config File

igraph init generates .igraph/config.json in the current directory. The graph database is stored at .igraph/igraph.db.

The config file has five sections:

{
  "embedding": { ... },    // Vectorization settings
  "llm": { ... },          // LLM summary settings
  "parser": { ... },       // Parser settings
  "retrieval": { ... },    // Retrieval settings
  "multimodal": { ... }    // Multimodal association settings
}

embedding — Vectorization

Field	Type	Default	Description
baseURL	string	http://localhost:8080/v1	Embedding API endpoint
model	string	bge-m3	Embedding model name
dimensions	number	1024	Vector dimensions
batchSize	number	32	Batch embedding size

llm — LLM Summaries

Field	Type	Default	Description
baseURL	string	https://api.openai.com/v1	LLM API endpoint
model	string	gpt-4o-mini	Symbol summary model
fileSummaryModel	string	gpt-4o	File summary model
temperature	number	0	Generation temperature
maxConcurrency	number	5	Max concurrent requests
promptVersion	string	v1.0	Prompt version identifier

parser — Parser

Field	Type	Default	Description
languages	string[]	["typescript", "javascript"]	Enabled languages
include	string[]	["**/*"]	File include globs
exclude	string[]	See below	File exclude globs

Default exclude patterns:

["node_modules/**", "dist/**", "**/*.test.*", "**/*.spec.*", "**/*.d.ts"]

retrieval — Retrieval

Field	Type	Default	Description
fileTopK	number	10	File-level retrieval count
nodeTopK	number	10	Symbol-level retrieval count
fallbackThreshold	number	0.75	Fallback threshold
graphHops	number	2	Graph expansion hops
fusion	string	rrf	Fusion strategy
rrfK	number	60	RRF fusion parameter
denseWeight	number	1.0	Dense channel weight
ftsWeight	number	1.0	FTS5 channel weight
resourceTopK	number	3	Independent resource retrieval count (searches resource_vectors directly)

multimodal — Multimodal Association

Field	Type	Default	Description
strongLinkThreshold	number	0.85	Strong association threshold
weakLinkThreshold	number	0.7	Weak association threshold
llmConfirmWeakLinks	boolean	false	Use LLM to confirm weak associations

Credential Management

Security Constraint

API Keys and other sensitive data must never be written to project-level config files (to prevent committing to git). Config validation actively scans for and rejects configs containing credential fields.

Recommended — Global Config

Use igraph config set once, shared across all projects:

igraph config set apiKey sk-xxx
igraph config set embedding.baseURL http://my-embedding:8080/v1
igraph config set llm.baseURL https://api.openai.com/v1

Global config is stored at ~/.igraph/config.json. View with igraph config list.

Environment Variables (Highest Priority)

Variable	Description	Required
IGRAPH_API_KEY	Shared key for LLM / Embedding	Required for online commands
IGRAPH_EMBEDDING_BASE_URL	Overrides embedding.baseURL	Optional
IGRAPH_LLM_BASE_URL	Overrides llm.baseURL	Optional

Priority

Environment variables > Global config (~/.igraph/config.json) > Project config (.igraph/config.json)

# Environment variable (temporary override)
export IGRAPH_API_KEY="sk-..."

Without an API Key:

• build / rebuild can run with --no-llm for heuristic fallback
• query / eval / serve automatically fall back to FTS5-only retrieval