It seems I cannot access the linked documentation pages. I will proceed with generating the brief based on the README content provided in the prompt.

Engram is an evidence-based learning engine designed to combat knowledge decay, particularly after interacting with AI explanations. It targets users who want to ensure long-term retention of learned material, acting as a personalized tutor, examiner, and scheduler. The primary benefit is transforming passive AI-generated understanding into durable, verifiable knowledge without requiring manual card creation or complex configuration.

How It Works

Engram orchestrates a learning loop involving several components. A Curriculum Architect breaks down topics into a first-principles concept map, identifying "threshold" concepts. A Tutor, leveraging an AI like Claude or Codex, guides the user through active recall: posing questions, prompting predictions, and requiring explanations back. An independent Assessor then grades these explanations blindly against a rubric, generating a "receipt" for each concept. Finally, a Scheduler, using the FSRS algorithm, analyzes these receipts to determine optimal future review dates, ensuring concepts are revisited just before they are forgotten. This approach prioritizes active generation and spaced retrieval, grounded in established learning science.

Quick Start & Requirements

Installation involves adding the plugin to Claude Code (claude plugin marketplace add nagisanzenin/engram and claude plugin install engram@engram) or OpenAI Codex (codex plugin marketplace add nagisanzenin/engram). It requires Python 3 (stdlib only) and the respective AI environment. Onboarding is minimal, with no configuration or accounts needed.

Highlighted Details

• Implements core learning science principles: retrieval practice, distributed practice, desirable difficulties, pretesting, and self-explanation.
• Utilizes FSRS-4.5 for personalized, spaced repetition scheduling based on recall performance.
• Features interactive HTML "explorables" for key concepts, offering manipulable models and embedded retrieval prompts.
• Employs blind assessment with verifiable "receipts" to ensure objective mastery tracking.
• Stores all learner state (model, graphs, receipts, artifacts) in human-readable JSON files locally at ~/.claude/learning/, ensuring data ownership and privacy.

Maintenance & Community

The project is maintained by "nagisanzenin." Specific community channels or detailed contributor information are not explicitly detailed in the provided README, but links to foundational documents, architecture, and roadmap are present.

Licensing & Compatibility

Engram is released under the MIT license. Its design emphasizes local data storage and processing, with artifacts being entirely offline, suggesting strong compatibility for local or air-gapped environments. The core engine is stdlib-only with no network code, except for the curriculum architect's initial web search for topic decomposition.

Limitations & Caveats

Engram is tightly integrated with Claude Code or OpenAI Codex, limiting its use to these platforms. The curriculum architect component performs web searches, meaning sensitive information should not be included in learning goals. While data is local, this also implies no built-in cloud synchronization.