memoryaf
Shared team memory for AI agents — MCP server, REST API, and dashboard backed by Antfly hybrid search.
What it is
memoryaf gives your AI agents persistent, searchable long-term memory — across sessions, teammates, and projects. It exposes an MCP server (for Claude Code, Cursor, and other MCP clients), a REST API (for any agent framework), and a web dashboard for browsing and managing memories. All backed by Antfly hybrid semantic + full-text search.
How it works
memoryaf supports three types of memory:
- Episodic — what happened. Chronological events: incidents, debugging sessions, decisions made in context.
- Semantic — what we know. Factual knowledge: architecture decisions, conventions, preferences.
- Procedural — how to do things. Workflow templates: runbooks, checklists, standard procedures.
On first startup, memoryaf auto-creates tables in Antfly with both a full-text index and an AKNN vector index. When you search, both indexes are queried and results are merged using reciprocal rank fusion, so you get the best of exact keyword matches and semantic similarity.
When Antfly Inference is available with the GLiNER2 model, memoryaf automatically extracts named entities (people, technologies, services, projects) from every memory and links them in an Antfly graph index — powering related-memory discovery and graph-expanded search.
Stack
- Runtime: Node.js + TypeScript
- MCP SDK:
@modelcontextprotocol/sdk - Database: Antfly via
@antfly/sdk - Dashboard: Next.js
- Embeddings: Antfly built-in (bge-small-en-v1.5, 384d), or Antfly Inference/Ollama for higher-quality models
- Entity Extraction: Antfly Inference with GLiNER2
- Search: Hybrid BM25 + AKNN vector with reciprocal rank fusion