API · Apr 2025 — May 2025

Search & Retrieval

Hybrid Search with Highlights, Facets, and Entity Expansion

Search & Retrieval

Full-text, vector, and graph search with strategy-aware retrieval policies.

Services

  • Text Search
  • Vector Search
  • Graph Expansion

Technologies

  • OpenSearch
  • Qdrant
  • Neo4j
  • SurrealDB

Software

  • FastAPI
  • Python
  • Semantic Kernel
  • Langchain

Client

CogMem Open Source

Introduction

CogMem provides a unified search API that combines full-text search with highlights, vector similarity search, and graph-based entity expansion. The active MemoryStrategy governs which retrieval mode is used — the system learns whether FTS, vector, graph, or hybrid search works best for your domain.

Challenges

  • Combining multiple search modalities into a unified API.
  • Adapting retrieval strategy based on domain characteristics.
  • Providing meaningful highlights and faceted filtering.
  • Expanding search results with related entities from the knowledge graph.

Solution

The search API resolves the active retrieval policy from the MemoryStrategy, then executes the appropriate search workflow. Results include highlighted snippets, faceted filters, and related entities. Outcome events are emitted after every search so the ACP can learn which retrieval approach works best.

Work Done

Hybrid Search

Combines full-text, vector, and graph search in a single API call.

Highlighted Results

Search results include highlighted snippets showing matching terms.

Faceted Filtering

Filter results by source, content type, and other facets.

Entity Expansion

Related entities from the knowledge graph are included in results.

Strategy-Aware

Retrieval mode adapts based on the active MemoryStrategy.

Outcome Tracking

Every search emits outcome events for the ACP to learn from.

Ready to Build Smarter Knowledge Systems?

Get started with CogMem and transform how your AI agents remember, organize, and evolve knowledge.

Get Started

Contact Us

Get in Touch

Have questions about CogMem? We'd love to hear from you.

Website

cogmem.ai