Hybrid Search

Hybrid Search: How the System "Thinks"

The MCP Local Memory Service uses a Hybrid Search Engine to ensure your AI Agent always finds the right information, even if you use different words or make typos.

🔍 How it Works

Every search result is the weighted blend of four signals:

1. Semantic similarity (40%) — meaning-based relevance using the all-MiniLM-L6-v2 model locally via Transformers.js. This lets the Agent understand that "database schema" is related to "migrations", even when the words don't match.
2. Keyword match (30%) — exact tokens and phrases found in stored text. A query for "auth" immediately finds content containing that exact term.
3. Recency (15%) — newer entries score higher; the signal halves every ~30 days.
4. Domain / workspace affinity (15%) — a boost when the memory's repository or folder matches your current working context (e.g. working in src/auth/ boosts memories scoped to the auth folder or repo).

The blend is computed as: finalScore = similarity·0.40 + keyword·0.30 + recency·0.15 + domain·0.15.

🧠 Smart Features

• Adaptive Thresholding: the strictness adapts to the size of the result set — lenient for small sets (0.10 for memories) so a fresh project still returns results, stricter for larger sets (0.40) to cut noise. If every candidate falls below the threshold, the single best match is still returned, so a cold start is never empty.
• Tech-Stack Affinity: pass current_tags (e.g. ["react", "laravel"]) to include memories tagged with those technologies from other projects. Your Agent's experience with a library in Project A follows to Project B.
• Conflict Prevention: storing a memory that contradicts an existing one (cosine similarity ≥ 0.85) is rejected with a MEMORY_CONFLICT error, keeping your knowledge base a single source of truth.

⚠️ Disclaimer

Semantic search performance depends on local CPU capabilities and the quality of the stored text. THE SOFTWARE IS PROVIDED "AS IS", without warranty of accuracy.

Related Pages

• Core Features — feature overview
• Tool Reference & Usage Guide — searching memories with memory-read
• Pencarian Hibrida — Bahasa Indonesia