Roadmap

Phase 1 — MCP Server MVP ✅ Shipped

The foundation: a working MCP server with full conversation storage and semantic search. Live at mcp.getengram.app.

• Monorepo setup (pnpm workspaces + Turborepo)
• Shared types, Zod schemas, and utilities
• D1 database schema + indexes
• Typed query helpers for all tables
• MCP server with core memory tools
• API key authentication (SHA-256 hashed)
• Sliding-window message chunking
• Embedding generation via Workers AI
• Vectorize semantic search with org-scoped filtering
• Tenant isolation at every layer
• Comprehensive test suite
• Deploy to Cloudflare Workers
• Production API key provisioning

Phase 2 — API and Operations ✅ Shipped

Management capabilities beyond the MCP protocol.

• Rate limiting per organization
• Usage tracking and metering
• Webhook notifications (conversation created, messages appended)
• Conversation exports
• API key management (create, list, revoke)
• Public REST API for data (append/search over HTTP without MCP)
• Bulk operations (bulk delete)

Phase 3 — Dashboard and Teams ✅ Shipped

A web interface for managing Engram without touching code.

• Dashboard UI
• OAuth 2.1 authorization server (self-serve app connections, e.g. ChatGPT)
• Team management (seats, billing, connected apps)
• API key management UI
• Connected-apps management (list + revoke)
• Conversation browser — view and search conversations in the UI
• Analytics — message volume, search patterns, active agents

Phase 4 — In progress

• Admin dashboard — business metrics and operational visibility
• ChatGPT plugin directory listing (public, no Developer mode required)
• Deterministic capture adapters for more hosts (Cursor hooks, etc.)
• Enterprise history backfill for ChatGPT Business/Enterprise workspaces

Future Considerations

• Async embedding pipeline — Move chunking/embedding to Cloudflare Queues to reduce append latency
• Configurable chunking — Let users tune window size and stride per organization
• Multiple embedding models — Support larger/multilingual models
• Cross-conversation search — Search with conversation-level context (not just chunk-level)
• Conversation branching — Fork conversations for A/B testing agent responses
• Retention policies — Automatic cleanup of old conversations by age or count