Beta AI MCP Knowledge Management

3ngram

Knowledge workspace for AI-native teams

A hosted knowledge workspace with semantic search, structured memory, and an MCP server that gives AI assistants persistent context across sessions.

Visit product
3ngram MCP
connected
▶ remember
topic: auth-architecture
type: decision
content: "Extracted shared auth into b3dmar-auth-core.
JWT + Argon2id + RBAC. Stateless, zero-opinion."
saved id: 2041
▶ recall "auth architecture"
0.94
Extracted shared auth into b3dmar-auth-core
decision · 2d ago
0.87
OAuth 2.0 with Streamable HTTP for MCP transport
context · 5d ago
0.81
Fail-open revocation mode for staging environments
pattern · 1w ago
▶
11 tools · pgvector · semantic search
3ngram.ai →

What is 3ngram?

3ngram is a knowledge workspace designed for people who work with AI assistants daily. It solves a fundamental problem: AI sessions are stateless. Every new conversation starts from zero.

3ngram gives your AI assistants persistent memory. Decisions you made last week, commitments you tracked yesterday, patterns you’ve noticed over months — all searchable, all available as context in your next session.

How it works

The core is an MCP server (Model Context Protocol) that exposes 11 tools to any compatible AI assistant:

  • remember — persist a decision, commitment, blocker, or context note
  • recall — retrieve memories by topic with semantic similarity search
  • search — full-text and vector search across your knowledge base
  • commitments / blockers / overdue — structured queries for accountability tracking
  • handoff — export session context as JSON for switching between AI providers

The web interface provides a dashboard for browsing, editing, and managing your knowledge base directly.

Architecture

  • Backend: FastAPI + PostgreSQL with pgvector for semantic search
  • Embeddings: OpenAI text-embedding-3-small via async batch processing
  • Auth: OAuth 2.0 with JWT tokens (powered by b3dmar-auth-core)
  • MCP transport: Streamable HTTP (SSE-based, compatible with Claude Code, Cursor, and other MCP clients)
  • Frontend: React + TypeScript with Cmd+K search, timeline views, and memory management
  • Hosting: Railway (backend + DB) + Vercel (frontend)

Status

Currently in beta with active daily use. Approaching public launch at v0.4.9. The MCP server is live at mcp.3ngram.ai and the web app at app.3ngram.ai.

All products