Home

Memongo platform overview

Memongo is a MongoDB-native long-term memory system for AI agents and apps. It stores conversations, facts, procedures, knowledge-base chunks, episodes, and graph relationships in MongoDB, then retrieves relevant context with vector search, full-text search, and hybrid ranking.

What it does

An AI agent talks to Memongo through an HTTP API, a TypeScript SDK, an MCP server, or AI SDK tool helpers. Memongo writes what it's told to remember, derives structured facts and procedures from raw events in the background, and answers search and recall calls with ranked, provenance-tagged results. The memory taxonomy organizes what's stored; the scope model controls who can see it.

The public repository ships a runnable product: an HTTP API, an MCP server, a TypeScript client, AI SDK tool helpers, a web console, docs, a Docker MongoDB setup, and release checks. Historical, experimental, and comparison material (other retrieval backends, abandoned prototypes) is intentionally kept out of this repo's supported surface — see CONTRIBUTING.md.

Who uses it

• Agent builders wire an AI agent to Memongo via packages/client (MemongoClient) or packages/tools (Vercel AI SDK / OpenAI tool helpers) so the agent can save and recall memory mid-conversation.
• MCP-compatible hosts (coding agents, chat clients) attach apps/mcp as a stdio MCP server and get the same operations as typed tools.
• Operators run apps/api against a MongoDB Atlas (cloud or Atlas Local Preview) cluster and watch it through apps/web, the operator console.

Quick links

• Architecture — how the API, bridge, engine, and MongoDB fit together
• Getting started — install, run MongoDB, start the API, add and search memory
• Glossary — Memongo-specific vocabulary
• Memory taxonomy — the six memory types and core operations
• Packages and Apps — the workspace map
• By the numbers — codebase size and activity snapshot

Repository layout

memongo/
  apps/
    api/          HTTP API server (Hono)
    mcp/          MCP server (stdio, calls the HTTP API)
    web/          Next.js operator console
    docs/         Public docs (Mintlify)
  packages/
    memory-engine/   Core MongoDB memory: embeddings, graph, episodes, search, KB, analytics
    memory-bridge/   Stable facade over the engine, used by apps/api
    memongo-memory/  Published `@memongo/memory` re-export barrel
    client/          TypeScript HTTP client SDK
    tools/           AI SDK tool helpers (Vercel AI SDK, OpenAI)
    lib/             Shared types, auth, and utilities (internal)
    pi-extension/    Pi coding-agent extension (additive semantic memory)
  docker/            Local MongoDB dev stack (Atlas Local Preview + mongot)
  scripts/           Benchmark, proof-pack, and release-gate scripts
  docs/              Platform docs, ADRs, and benchmark evidence

Product framework contract

Memongo's public framework contract, repeated throughout this wiki:

• Memory taxonomy: episodic events, semantic facts, procedural playbooks, profile preferences, workspace knowledge, and provenance.
• Core operations: recall, context bundles, remember, update, forget, feedback, and trace.
• Scope model: session, user, agent, workspace, tenant, and global.
• Safety model: read by default; write only on explicit user, app, operator, test, or import intent.

---

Generated by Factory