The problem

AI assistants are stateless. Every conversation starts from scratch. You re-explain your project structure, remind it of decisions you made last week, and repeat context that should already be known. The more you use AI, the more this friction compounds.

What 3ngram does

3ngram gives your AI assistant a persistent memory. When you make a decision, it gets stored. When you commit to a deadline, it gets tracked. When you start a new session tomorrow, your AI already knows what happened today.

Think of it as a Chief of Staff that sits between you and your AI tools — briefing them on context, flagging overdue commitments, and making sure nothing falls through the cracks.

How it works

3ngram connects to your AI assistant through MCP (Model Context Protocol) — the open standard for extending AI tools. Once connected, your assistant gets access to 30+ tools:

Remember and recall — store decisions, patterns, and context with semantic search
Commitments and blockers — structured accountability tracking with due dates
Morning briefings — auto-generated session context from recent activity
Triage and reclassify — Chief of Staff workflows for keeping your knowledge base clean
Handoffs — export context as JSON when switching between AI providers

The web dashboard at app.3ngram.ai lets you browse, edit, and manage your memories directly.

Under the hood

Backend: FastAPI with PostgreSQL + pgvector for semantic similarity search
Embeddings: OpenAI text-embedding-3-small, processed in async batches
Auth: OAuth 2.0 with JWT tokens, powered by b3dmar-auth-core
MCP transport: Streamable HTTP (SSE-based), compatible with Claude Code, Cursor, Windsurf, and other MCP clients
Frontend: React + TypeScript — Cmd+K search, timeline views, memory management
Infrastructure: Railway (API + database) + Vercel (frontend)

Status

Live and in daily use. Web app at app.3ngram.ai, docs at docs.3ngram.ai. Preparing for public launch with team workspaces and open-source MCP server release.