Introduction

Engrams is a governance-aware, context-intelligent memory platform built on the Model Context Protocol (MCP). It transforms how AI agents understand and work with your projects by providing structured memory, intelligent context retrieval, and visual knowledge exploration.

🚀 Just want to dive in? Skip straight to Installation or the Quick Start guide.

What problem does Engrams solve?

AI coding assistants are powerful — but they forget everything between sessions. You end up re-explaining your tech stack, your architectural decisions, and your coding conventions in every single conversation. The larger your project grows, the worse this gets.

Engrams gives your AI a persistent, queryable project memory. Instead of loading your entire codebase into the context window, Engrams surfaces only what's relevant to the current task — automatically.

Stored knowledge types

TypeDescriptionExample
Decisions Architectural & implementation choices with rationale "Use PostgreSQL — we need ACID guarantees"
Progress Tasks, sub-tasks, and their statuses TODO / IN_PROGRESS / DONE
Patterns Recurring design or architectural patterns "JWT validation middleware"
Context Project goals, current focus, team agreements Product context & active context
Custom Data Glossaries, specs, any structured project knowledge Domain glossary terms

Key capabilities

  • Semantic search — find context by meaning, not keyword
  • Codebase bindings — link decisions to file paths; context follows the code
  • Context budgeting — token-aware selection keeps costs low as your knowledge base grows
  • Team governance — enforce team-level architectural mandates with a two-scope hierarchy
  • Knowledge Dashboard — browser-based explorer with D3 graph visualization
  • Export / Import — markdown export for version control and team sharing

Architecture overview

Engrams runs as an MCP server alongside your AI tool. Your IDE extension (Roo Code, Cline, Cursor, Windsurf, etc.) communicates with Engrams over stdio or HTTP. Each project gets its own isolated SQLite database in a context_portal/ directory at the project root.

You
↔
AI Agent
(Roo, Cline, Cursor…)
↔
Engrams MCP
(uvx engrams-mcp)
↔
SQLite DB
context_portal/context.db

Next steps