Repolode
Submit
nagisanzenin

nagisanzenin/engram

A learning engine for Claude Code and other agent platforms that tutors you, blind-grades your free recall, and schedules reviews with FSRS so you retain what you learn.

Open-source alternative to:Anki

View on GitHub
Stars
1,474
Forks
108
Open issues
5
Last commit
2 months ago

What is it?

Engram is a learning system for the human, not a memory tool for the AI agent. You type a command such as /learn with a topic, and the agent builds a first-principles concept map, asks you to predict and explain before it teaches, and then grades your answers in a separate assessor agent that never sees the lesson. Each grade is written to a local receipt file, and a deterministic Python scheduler built on FSRS-4.5 sets the next review date for each concept.

Day-to-day use centers on short sessions: /review runs two to four minutes of free recall across due concepts, and /coach shows retention stats, calibration and a local HTML dashboard. Learner state is stored as plain JSON under ~/.claude/learning/, and the engine is stdlib-only Python with no network code. The project is distributed as a Claude Code plugin and also installs on Codex, OpenCode, Hermes Agent, OpenClaw, Pi and several other platforms.

The repository publishes a gold set of graded answers and an audit command, so the grader's leniency can be checked. It also provides an opt-in export that writes a file for review without sending anything.

Key features

  • 01Tutor-style lessons that ask you to predict and explain before the answer is given.
  • 02Blind grading of free recall by a separate assessor agent, with each grade saved as a receipt.
  • 03FSRS-4.5 scheduling that sets review dates from your own review history.
  • 04Two- to four-minute /review sessions that interleave due concepts across topics.
  • 05Local, human-readable JSON state with no network code in the engine.

Use cases

  • Learning a technical topic such as machine learning or Rust and keeping it past the first month.
  • Studying STEM material that involves procedures, with worked examples and fresh practice problems.
  • Replacing self-graded flashcards with a schedule built from a tutor-led lesson.
  • Running short weekly review sessions alongside other coding work in Claude Code.

Quick start

  1. 1

    Add the Engram marketplace in Claude Code

    claude plugin marketplace add nagisanzenin/engram
  2. 2

    Install the Engram plugin

    claude plugin install engram@engram
  3. 3

    Start learning a topic inside your coding assistant

    /learn kalman filters

Things to know

  • Requires python3 to be installed on your machine.
  • Built for Claude Code first; other platforms have partial caveats such as a pending nudge on Antigravity and a developer-preview status on DeepSeek Harness.

In collections

Claude Code tools, skills & pluginsLearning & education tools

Topics

#learning-science#spaced-repetition#fsrs#claude-code-plugin#education#active-recall#claude-code#learning

FAQ

What is engram?
A learning engine for Claude Code and other agent platforms that tutors you, blind-grades your free recall, and schedules reviews with FSRS so you retain what you learn. Engram is a learning system for the human, not a memory tool for the AI agent. You type a command such as /learn with a topic, and the agent builds a first-principles concept map, asks you to predict and explain before it teaches, and then grades your answers in a separate assessor agent that never sees the lesson. Each grade is written to a local receipt file, and a deterministic Python scheduler built on FSRS-4.5 sets the next review date for each concept.
Is engram free to use? What license does it have?
engram is open source under the MIT license. Check the license terms before using it in commercial products.
Is engram still maintained?
Updated within 6 months. The last commit to engram was on Aug 27, 2026.
What can I use engram for?
Learning a technical topic such as machine learning or Rust and keeping it past the first month. Studying STEM material that involves procedures, with worked examples and fresh practice problems. Replacing self-graded flashcards with a schedule built from a tutor-led lesson. Running short weekly review sessions alongside other coding work in Claude Code.
How do I install engram?
Add the Engram marketplace in Claude Code: claude plugin marketplace add nagisanzenin/engram Install the Engram plugin: claude plugin install engram@engram Start learning a topic inside your coding assistant: /learn kalman filters
Where can I get engram?
The source code and installation instructions are on GitHub: https://github.com/nagisanzenin/engram

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects