Konnektr offers a hybrid graph‑plus‑vector database that serves as semantic memory for AI agents, storing validated property graphs with embedded vectors in PostgreSQL enhanced by Apache AGE and pgvector. It lets developers define schemas for data integrity and run combined Cypher and vector similarity queries, enabling agents to retrieve, traverse, and update structured knowledge through natural‑language tool calls via its Model Context Protocol.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents that rely solely on vector-based retrieval lack explicit relationships, schema enforcement, and the ability to traverse hierarchical dependencies, leading to inconsistent or low-quality context as knowledge bases expand.
Solution
Konnektr provides a hybrid graph‑plus‑vector database that serves as semantic memory for AI agents. By storing validated property graphs with embedded vectors in PostgreSQL enhanced with Apache AGE and pgvector, it enables agents to query both semantic similarity and explicit relationships in a single operation. Users define ontologies and schemas, ensuring data integrity at write time, while agents can create, update, and traverse entities through natural‑language tool calls. The platform exposes a Model Context Protocol (MCP) that integrates with LLMs such as Claude or Cursor, allowing agents to perform searches, execute Cypher traversals, and write validated entities via REST, SDK, or MCP interfaces. This structured memory improves context relevance, supports dependency analysis, and scales as the knowledge graph grows.
Target Audience
Primary customers are developers and enterprises building AI agents, digital twin systems, or knowledge‑intensive applications that require both relational context and semantic search capabilities.
Features
- PostgreSQL backend augmented with Apache AGE for property graph capabilities and pgvector for high‑dimensional embeddings
- Schema definition and validation to prevent contradictory or malformed data during writes
- Hybrid query engine that combines Cypher graph traversal with vector similarity ranking in a single statement
- Model Context Protocol (MCP) exposing graph operations as LLM‑compatible tools (search, query, create/replace)
- Multiple ingestion methods including SDK, REST API, and bulk MCP ingestion pipelines
- Compatibility with major LLMs (e.g., Claude, Cursor) for seamless conversational interaction with the knowledge graph