Skip to content

Home

Heta — Agent-Oriented Knowledge Management Platform

Agent-Oriented Knowledge Management Platform
Multi-database orchestration · Dual-mode memory · Structured knowledge generation

What is Heta?

Heta is an all-in-one knowledge infrastructure for AI agents. It integrates multiple underlying databases to help agents do three things: acquire external knowledge, accumulate their own memory, and reason and generate based on that knowledge.

  • HetaDB — Unified management across multiple databases for smarter external knowledge integration. No need to worry about what format the knowledge comes from or which database it should go into.
  • HetaMem — Dual-mode memory: fast episodic recall via vector search (MemoryVG), and a long-term knowledge graph that evolves continuously with the agent (MemoryKB).
  • HetaGen — Synthesizes and expands upon existing knowledge bases to generate higher-value structured content.

Key Features

HetaDB

  • Multi-format file parsing via MinerU and other tools (PDF, HTML, images, sheets, Markdown, Office file formats, archives, etc.)
  • Parsed knowledge is automatically distributed across the appropriate databases — no manual storage decisions required
  • Multiple query strategies for different scenarios: naive (vector search)， rerank (BM25 + vector + cross-encoder)， rewriter (query pptimization)， multihop (ReAct)， direct (direct query)
  • Source document tracing to support agentic search workflows

HetaMem

  • MemoryVG — Lightweight memory for high-frequency, fragmented information: quickly stores and retrieves conversation content, user preferences, and contextual facts; full CRUD + history audit
  • MemoryKB — Empowers AI agents with continuous, multimodal memory by combining hierarchical knowledge graphs with fast parametric recall, enabling long-term context understanding and personalized, adaptive interactions.

HetaGen (early stage)

  • Generates structured tabular data from knowledge bases and supports Text-to-SQL queries on the generated tables
  • Automatically construct hierarchical knowledge structures (tag trees) from topics

MCP Integration

HetaDB and HetaMem expose optional MCP servers (ports 8012 / 8011) for direct integration with Claude Desktop, Cursor, and other MCP-compatible clients.