Project overview
Engram is an AI learning tool for people, with a Python core engine and plugins or skills for assistants such as Claude Code, OpenAI Codex, and OpenCode. It helps people learn and review knowledge rather than adding long-term memory to an AI assistant.
The project connects curriculum planning, conversational tutoring, independent assessment, and FSRS spaced review. FSRS uses review history to schedule the next review. Learning state is stored in readable JSON files, by default at `~/.claude/learning/`. Interactive learning materials and local dashboards are stored in its `artifacts/` folder.
Use cases
Engram suits self-directed learners who want to practice after an AI explanation, check their understanding, and return for scheduled reviews. README examples include programming, mathematics, statistics, music theory, and history.
For conceptual knowledge, learners predict, answer, and explain ideas in their own words. For procedural material, they study examples, complete missing steps, and then solve fresh variants independently. Learning maps and a local dashboard help users inspect their progress.
What it does and problems it solves
Engram addresses learning that stops at reading explanations, without independent answers or follow-up review, through these steps:
- Break topics into a learning route based on prerequisite relationships.
- Ask learners to attempt answers before receiving hints and explanations.
- Use an assessor in a separate context to grade against a rubric and the learner's answers, saving assessment records.
- Schedule reviews from those records and display progress, retention information, and upcoming reviews.
- Generate interactive materials with prediction and explanation prompts for concepts that benefit from manipulation.
Local storage does not mean the entire AI learning process is offline. The README says curriculum planning may search the web using the supplied topic and goal. Avoid including sensitive information in the goal, or request an offline map. The supplied material does not state the costs of connected AI platforms and does not establish a guarantee of learning outcomes.
How to use
1. Install Engram for your chosen platform using the next section, then open that AI assistant.
2. In Claude Code or OpenCode, enter `/learn kalman filters` to start the example topic, or replace the topic with something you want to learn. Codex uses `$learn`; consult the original repository for command forms on other platforms.
3. Answer the assistant's questions about your goals and background, complete the pretest, and attempt predictions, answers, and explanations in your own words during lessons. The system then saves records and schedules reviews based on independent assessment.
4. Use `/review` for due reviews, or `$review` in Codex. Conceptual material primarily uses recall, while procedural material uses fresh problem variants.
5. Use `/coach` to inspect learning statistics and discuss review scheduling, or `$coach` in Codex. The README also provides `/coach dashboard` to generate a local dashboard.
Installation
1. Prepare your chosen AI assistant and make sure `python3` is available. The README says the core engine uses only the Python standard library, but does not specify a minimum Python version or provide complete installation tutorials for the assistants themselves. Consult the original repository and the relevant platform documentation if these prerequisites are missing.
2. For Claude Code, run `claude plugin marketplace add nagisanzenin/engram` and then `claude plugin install engram@engram` in a terminal.
3. For OpenAI Codex, instead run `codex plugin marketplace add nagisanzenin/engram` and then `codex plugin add engram@engram`. Consult `INSTALL-CODEX.md` in the original repository for further configuration.
4. For OpenCode, add `opencode-engram-learning` to the `plugin` array in your existing `opencode.json`. The README allows global or project-level configuration. Consult `INSTALL-OPENCODE-V2.md` for OpenCode 2.0 beta details.
5. For other platforms, follow their installation documents in the original repository. If cloning or downloading is required, obtain the corresponding clone address or download entry from the original repository below this page. The supplied material does not include complete configuration steps for every platform, so the commands above should not be applied to all of them.
This page lists only the repository link. A runnable package and the core workflows have not been validated. These instructions come from the README and do not indicate that installation or runtime verification has been completed.
Source and usage notes
- Original repository: https://github.com/nagisanzenin/engram
- Source revision: 0590eb8f008680acbfb5aaa4dbc73bf63ca643dd
This listing links to the upstream repository. It does not provide a verified download package, and the complete workflow has not been independently validated.
This description was organized and translated with AI from public upstream material. Refer to the original repository for capabilities, setup requirements, costs, and license terms. 
