Configure Intelligence

MemoFS uses a 4-role intelligence model to power semantic search, knowledge graph extraction, and memory consolidation. Every role features a zero-dependency deterministic local fallback that runs without requiring third-party API keys or external model downloads.

The 4-Role Intelligence Model

1. MemoryEmbedder (Vector Embeddings)

Computes dense vector representations for semantic similarity scoring during recall:

• Local Fallback: When omitted, recall operates in lexical-only mode using BM25 token frequencies and fuzzy edit distance.
• Provider Adapters:
  • @memofs/adapter-openai (text-embedding-3-small, text-embedding-3-large)
  • @memofs/adapter-voyage (voyage-4, voyage-3, voyage-code-3)
  • @memofs/adapter-transformers (Local in-process ONNX embeddings)

2. Reranker (Candidate Rescoring)

Re-scores and sorts retrieved candidates before returning search results:

• Local Fallback: DeterministicFallbackReranker calculates lexical token-overlap between query and candidate text.
• Provider Adapters:
  • @memofs/adapter-voyage (rerank-2.5-lite)

3. Extractor (Graph Entity & Edge Extraction)

Extracts entity vertices and relationship edges from prose when memories are written:

• Local Fallback: Built-in rule-based extractor (createRuleBasedExtractor) evaluating 7 linguistic structural patterns (depends on, uses, supersedes, prefer, etc.).
• Provider Adapters:
  • @memofs/adapter-workers-ai (createWorkersAiExtractor)

4. LlmClient (Generative Intelligence)

Provides completion capabilities for LLM-enhanced prompt briefing generation and knowledge graph consolidation:

• Local Fallback: Heuristic and regex-based strategist pipeline without LLM roundtrips.
• Custom Adapters: Any provider implementing the core LlmClient contract (name, complete()).

Configuration Example

import { createHostedRuntime } from "@memofs/server";
import { InMemoryMemoryStore, createRuleBasedExtractor } from "@memofs/core";
import { createOpenAIEmbedder } from "@memofs/adapter-openai";
import { createVoyageReranker } from "@memofs/adapter-voyage";

const memofs = createHostedRuntime({
  store: new InMemoryMemoryStore(),
  projectId: "intelligence-demo",

  // 1. Vector embedder for semantic search
  embedder: createOpenAIEmbedder({
    apiKey: process.env.OPENAI_API_KEY,
    model: "text-embedding-3-small",
  }),

  // 2. Semantic reranker for high-precision recall
  reranker: createVoyageReranker({
    apiKey: process.env.VOYAGE_API_KEY,
  }),

  // 3. Knowledge graph extractor
  extractor: createRuleBasedExtractor(),
});