Skip to content

configuration.md

Latest commit

 

History

History
142 lines (110 loc) · 4.41 KB

configuration.md

File metadata and controls

142 lines (110 loc) · 4.41 KB
Raw

Configuration

GsRag Constructor Options

interface GsRagConfig {
  /** LLM and embedding providers */
  providers: LightRAGProviderRegistry;

  /** Storage registry (pre-built or custom) */
  storages?: ExtendedStorageRegistry;

  /** Storage factory (for multi-workspace) */
  storageFactory?: (ctx: { workspace: string }) => ExtendedStorageRegistry;

  /** Workspace name (default: "default") */
  workspace?: string;

  /** Document readers (auto-created if omitted) */
  readers?: DocumentReader[];

  /** Password for encrypted PDFs */
  pdfDecryptPassword?: string;

  /** Document processing config */
  document?: DocumentConfig;

  /** Query config */
  query?: QueryConfig;

  /** Tokenizer model name */
  tokenizerModel?: string;

  /** Max concurrent file reads (default: 4) */
  maxConcurrentFileReads?: number;
}

DocumentConfig

interface DocumentConfig {
  /** Chunk size in tokens (default: 1200) */
  chunkTokenSize?: number;

  /** Overlap between chunks in tokens (default: 100) */
  chunkOverlapTokenSize?: number;

  /** Max documents processed in parallel (default: 4) */
  maxConcurrentDocuments?: number;

  /** Split by character delimiter instead of token count */
  splitByCharacter?: string;

  /** Don't split large sections further when splitByCharacter is set */
  splitByCharacterOnly?: boolean;

  /** Max entity name length */
  maxEntityNameLength?: number;

  /** Language for extraction prompts */
  language?: string;

  /** Enable LLM response cache for entity extraction */
  enableLlmCacheForEntityExtract?: boolean;

  /** Max concurrent LLM calls during extraction */
  llmModelMaxAsync?: number;
}

QueryConfig

interface QueryConfig {
  /** Enable LLM response cache for queries */
  enableLlmCache?: boolean;
}

QueryParam Defaults

QueryParam can be constructed with partial overrides. All fields have sensible defaults:

Field Default Env Override Description
mode "mix" — Query mode
onlyNeedContext false — Return raw context only (no LLM)
onlyNeedPrompt false — Return rendered prompt only
responseType "Multiple Paragraphs" — LLM response format
stream false — Stream response via AsyncIterable
topK 40 TOP_K Top entities/relations to retrieve
chunkTopK 20 CHUNK_TOP_K Top chunks to retrieve
maxEntityTokens 6000 MAX_ENTITY_TOKENS Entity context token limit
maxRelationTokens 8000 MAX_RELATION_TOKENS Relation context token limit
maxTotalTokens 30000 MAX_TOTAL_TOKENS Total context token limit
hlKeywords [] — Pre-set high-level keywords
llKeywords [] — Pre-set low-level keywords
conversationHistory [] — Chat history for multi-turn
historyTurns 0 HISTORY_TURNS Turns of history to include
userPrompt "" — Additional LLM instructions
enableRerank true RERANK_BY_DEFAULT Enable reranking
includeReferences false — Include references in response
useLlmKeywordExtraction true — Use LLM for keyword extraction

Environment Variables

Variable Default Used By
OPENAI_API_KEY — OpenAI provider
ANTHROPIC_API_KEY — Anthropic provider
GOOGLE_API_KEY — Gemini provider
OLLAMA_HOST http://localhost:11434 Ollama provider
DATABASE_URL — Postgres storage
POSTGRES_URL — Postgres storage (fallback)
TOP_K 40 QueryParam
CHUNK_TOP_K 20 QueryParam
MAX_ENTITY_TOKENS 6000 QueryParam
MAX_RELATION_TOKENS 8000 QueryParam
MAX_TOTAL_TOKENS 30000 QueryParam
HISTORY_TURNS 0 QueryParam
RERANK_BY_DEFAULT true QueryParam
WORKSPACE default Workspace selection
LIGHTRAG_FAST_LOCAL_PIPELINE — Skip extraction + KG merge

Provider Configuration Options

All completion providers accept these common options:

{
  model: string;
  apiKey?: string;         // or set via env var
  baseURL?: string;        // custom API endpoint
  timeoutMs?: number;      // default: 30_000
  maxRetries?: number;     // default: 3
  temperature?: number;
  maxTokens?: number;
}

See Providers for provider-specific options.