git404hub

patdolitse/engram

★ 56PythonAudience · developerComplexity · 2/5ActiveLicenseSetup · easy

tl;dr

Local MCP server that stores developer identity, preferences, and lessons in plain JSON and Markdown so AI coding tools can read shared personal context across sessions.

vibe map

mindmap
  root((engram))
    Inputs
      Identity facts
      Lessons learned
      Session summaries
    Outputs
      MCP responses
      Identity card markdown
      Knowledge health score
    Use Cases
      Cross tool memory
      Code review standards
      Project context priming
    Tech Stack
      Python
      MCP
      JSON
      Markdown
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

what people make with this

VIBE 1

Keep one set of personal coding standards that Claude Code, Cursor, and Codex all read

VIBE 2

Mine lessons and decisions from a pasted session summary into structured notes

VIBE 3

Generate an identity card to paste into AI tools that do not support MCP

VIBE 4

Audit stale knowledge entries and merge duplicate notes across projects

stack

PythonMCPJSONMarkdown

setup vibes

Difficulty · easy time til it works · 30min

Each MCP client needs its own config snippet pasted in and a restart before engram is active.

Apache 2.0 license, free to use and modify commercially with a patent grant and attribution.

in plain english

Engram is a small local program that stores facts about you, the developer, and serves them to AI coding tools through the Model Context Protocol, or MCP. MCP is an open standard that AI assistants like Claude Code, Codex, and Cursor use to read data from external sources. The pitch is that every time you start a new chat with an AI coding assistant, it forgets your communication preferences, your code review standards, your past mistakes, and the reasoning behind decisions you made months ago. Engram keeps that information in plain files on your machine so any MCP-compatible tool can read the same context. The data lives in a folder called ~/.engram/ as JSON and Markdown files that you can open, edit, back up, or migrate by hand. The categories the project tracks are identity, quality standards, preferences, trust boundaries, project snapshots, lessons learned, key decisions, and domain knowledge. The README is careful to distinguish this from agent memory tools like Mem0, Zep, or Letta, which store task-level session history. Engram is positioned as a layer above that: it stores who you are as a person, rather than what a particular agent did in a particular task. Beyond storage, the project exposes a handful of active functions over MCP. get_knowledge_inheritance pulls a starter pack of relevant lessons when you describe a new project. extract_session_insights takes a pasted session summary and mines lessons and decisions out of it automatically. get_identity_card emits a Markdown summary you can paste into AI tools that do not speak MCP, such as ChatGPT or Gemini. get_knowledge_overview surfaces stale entries and gives a health score, and find_similar_knowledge and link_knowledge help you merge duplicates and build links between related notes. Installation is a Python package called piia-engram, installable with pip, followed by a setup script and a small JSON snippet that you paste into your AI tool's MCP configuration. After restarting the client, a new session will automatically call get_user_context. There is an engram doctor command that can check Claude Code, Cursor, and Claude Desktop for stale MCP configs and auto-repair them. The license is Apache 2.0.

prompts (copy fr)

prompt 1
Install engram and wire it into my Claude Code MCP config so a new session auto-loads my preferences
prompt 2
Show me the JSON schema engram uses under ~/.engram and how to bulk import notes from an existing markdown folder
prompt 3
Add a new MCP function to engram that returns only the lessons tagged with a specific language
prompt 4
Run engram doctor and walk me through what it changed in my Cursor and Claude Desktop configs
prompt 5
Compare engram to Mem0 and explain how I would use them together without overlap
peek the repo → explain another one

← patdolitse on gitmyhub — every repo by this author, as a profile.

double-check against the repo, no cap.