Adaptive Practice

Adaptive Practice is an Obsidian community plugin that helps you practice and retain knowledge from your notes using spaced repetition and adaptive scheduling.

Features

• Generate adaptive practice questions directly from Obsidian notes and PDFs.
• Maintain a lightweight vault skeleton index with frontmatter, tags, links, headings, file stats, and embedded media references.
• Read selected notes more deeply at generation time, including frontmatter, outline, bounded section excerpts, and supported attachments.
• Daily practice reminders with a spaced, skill-aware topic queue stored in plugin data.
• Fluency-aware review that uses correctness, skips, and response time to detect fragile recall.
• A dashboard view for streak, due notes, scan status, and one-click daily practice.
• Multi-provider BYOK support for Gemini, Anthropic, OpenAI, DeepSeek, Qwen, OpenRouter, local Ollama, Claude Code and Codex subscriptions (via the native engine), and custom OpenAI-compatible endpoints.
• Obsidian-native Markdown output with LaTeX and code block guidance in the generation prompt.

Model providers

Open Settings → Adaptive Practice and choose a provider, then select or create the Obsidian secret that stores that provider's API key. API keys are stored through Obsidian secret storage. Each provider keeps its own configurable secret name, so switching between Gemini, Anthropic, OpenAI-compatible routes, etc. does not overwrite the key slot you chose for another provider.

Provider	Default endpoint/model	Notes
Gemini	gemini-3.5-flash	Supports image and PDF attachments through the Gemini API. Change the model in settings if your account uses a different Gemini model.
Anthropic	https://api.anthropic.com/v1/messages, claude-sonnet-4-6	Supports image and PDF attachments through the Messages API. Change the model in settings if your account uses a different Claude model.
OpenAI	https://api.openai.com/v1/responses, gpt-5.5	Uses the Responses API with structured JSON output. Change the model in settings if your account uses a different OpenAI model.
DeepSeek	https://api.deepseek.com/chat/completions, deepseek-v4-flash	Uses OpenAI-compatible chat completions with JSON object mode. Change the model in settings if you want the pro route.
Qwen	https://dashscope-intl.aliyuncs.com/compatible-mode/v1/chat/completions, qwen3.7-plus	Uses Alibaba Cloud Model Studio's OpenAI-compatible interface. Change the model in settings if your region or account uses another Qwen route.
OpenRouter	https://openrouter.ai/api/v1/chat/completions, openai/gpt-5.4-mini	Uses OpenRouter's OpenAI-compatible route. Change the model to any route you have access to.
Claude Code (subscription)	sonnet	Runs on your Claude subscription through the Whetstone native engine (desktop, CLI installed and signed in). Model aliases: sonnet, opus, fable.
Codex CLI (subscription)	terra	Runs on your ChatGPT subscription through the Whetstone native engine (desktop, CLI installed and signed in). Model aliases: terra, sol, luna.
Ollama (local)	http://localhost:11434/v1/chat/completions, llama3.1	Runs against your local Ollama server. No API key, free, fully offline. Pick any pulled model in settings.
OpenAI-compatible	http://localhost:1234/v1/chat/completions	For LM Studio, local gateways, or other compatible servers. Local endpoints may omit an API key.

Image attachments are sent only when the selected provider is configured to support vision input. PDF topics generate through the native engine and require Anthropic or Gemini (the providers that read documents natively); the engine refuses other providers with a clear message. Daily practice skips incompatible PDF topics and uses compatible due notes when available. Standalone PDF topics are capped at 10 MB per file before upload.

Daily practice

Use the Open dashboard command or the "brain" ribbon icon to open the Adaptive Practice dashboard. The dashboard shows your current streak, how many notes are ready for review, the last vault scan, and the selected daily review topics. From there you can start the daily session, choose notes manually, or rescan the vault skeleton.

When choosing notes manually, the picker uses practice memory to sort due and low-skill notes first. You can search by title/path, filter to due/new/low-skill/PDF notes, and select all currently visible matches without rendering the whole vault at once.

The scan is incremental: unchanged notes reuse their previous skeleton entry, while changed notes refresh metadata from Obsidian's cache. Scans yield back to Obsidian between batches so large vaults do not freeze the UI during startup or scheduled rescans. This keeps daily topic selection cheap enough for large vaults while still letting generation read the selected notes in detail. For very large notes, prompts keep the outline and sample representative section excerpts so a single clipped or encyclopedic note does not consume the whole context window.

After a session, the scheduler updates each note with recent accuracy, skipped answers, rolling average time, and a fluency score. Slow or skipped recall shortens the next interval and can bring a note forward in the daily queue even when the overall skill score is not low.

Adaptive Practice also tracks generated sourceSubtopics per note. Future prompts receive a compact subtopic memory block so the model can avoid already-mastered subtopics and revisit weak or due ones without rereading the full practice log.

Daily sessions also adjust the generated question count before calling the model: fragile recall gets a shorter warm-up, balanced review keeps your configured count, and strong recent accuracy/fluency earns a small stretch. This keeps token use bounded while nudging the session toward flow instead of a fixed-size batch every day.

---

Requirements

• Obsidian v1.11.4 or higher (desktop or mobile; desktop recommended for initial install).
• Node.js (LTS, v18+ recommended) for local builds.
• Git (for install via cloning).

Security disclosure

On desktop, this plugin runs a native helper program (the open-source Whetstone engine) as a separate process to generate and verify practice questions. That is why plugin scanners report shell or process execution.

What it actually does:

• The engine is downloaded once from a pinned GitHub release, and its SHA-256 checksum is verified before it is ever executed. A download that does not match the checksum carried in the plugin is discarded.
• The plugin only ever launches a fixed binary with fixed arguments (serve, --version). It never builds shell command strings and never runs with a shell.
• You can point the plugin at your own engine binary in settings, or build it yourself from the AGPL source.
• If you select the Claude Code or Codex provider, the engine invokes that CLI on your machine, using your existing login.
• On mobile, none of this applies and no processes are started.

---

Installation

You can install this plugin in two main ways:

1. Via Git clone + npm install (manual install)
2. Via the BRAT plugin (recommended for tracking this repo)

1. Manual Installation (Git clone + npm)

This method is best if you want to develop, tweak, or inspect the source.

1. Clone the repository

   cd /path/to/your/vault/.obsidian/plugins
   git clone https://github.com/anupchavan/adaptive-practice.git
   cd adaptive-practice
   

2. Install dependencies

   npm install
   

3. Build the plugin

   For a one-time production build:

   npm run build
   

   Or for development with watch:

   npm run dev
   

   After a successful build, you should have:

   • main.js
   • manifest.json
   • (optionally) styles.css

   at the root of the adaptive-practice plugin folder.

4. Enable the plugin in Obsidian

   • Open Obsidian.
   • Go to Settings → Community plugins.
   • Make sure Safe mode is turned off.
   • Select Browse, then Installed plugins.
   • Find Adaptive Practice and toggle it on.

---

2. Installation via BRAT (Beta Reviewers Auto-update Tester)

If you’d like Obsidian to automatically keep this plugin updated from a GitHub repo, you can use the BRAT plugin.

1. Install BRAT

   • In Obsidian, go to Settings → Community plugins.
   • Select Browse.
   • Search for “BRAT” (Beta Reviewers Auto-update Tester).
   • Install and enable BRAT.

2. Add this plugin as a beta plugin in BRAT

   • Open Command palette (Cmd/Ctrl+P) and run:

     • “BRAT: Add a beta plugin for testing”

   • When prompted for the GitHub repository, enter:

     anupchavan/adaptive-practice
     

3. Enable the plugin

   • BRAT will download the plugin into your vault.
   • Go to Settings → Community plugins → Installed plugins.
   • Enable Adaptive Practice.

4. Auto-updates

   • BRAT can periodically check for updates from the GitHub repo and update the installed plugin.
   • Configure BRAT’s update settings from Settings → Community plugins → BRAT as desired.

---

Development

If you want to modify the plugin:

1. Make sure you are in the plugin directory:

   cd /path/to/your/vault/.obsidian/plugins/adaptive-practice
   

2. Run the dev build:

   npm run dev
   

3. After each change is built, reload Obsidian (or use “Reload app without saving” from the Command palette) to see the changes.

---

Support & Feedback

• For bugs or feature requests, please open an issue on the GitHub repository.
• If you find this plugin useful, consider starring the repo to show support.

---