AI Organizer for Obsidian

A note-aware AI workspace for writing, reading, image understanding, and knowledge-base maintenance.

Why AI Organizer · Interface · Installation · Configuration · Feature Reference · Privacy

AI Organizer brings AI into the note itself. Select a passage to translate or rewrite it, ask questions with the current note and its images as context, keep anchored reading notes outside the Markdown body, and maintain a vault with formatting, metadata, links, inbox, attachment, and batch tools.

It supports remote APIs and local OpenAI-compatible services. Text and vision models are configured separately, while built-in OCR provides a fallback when a vision model is unavailable.

Why AI Organizer

Interface

Context that follows your work

The sidebar can use the whole note or a stable selection snapshot. It resolves images referenced by that context, analyzes them through the active vision model or OCR fallback, and passes a concise result to the text model. The ring beside the composer estimates usage against the context-window value configured for the active model.

A workbench that gets out of the way

The workbench groups note-level actions without permanently taking conversation space. Collapse it when you only need chat; expand it for formatting, annotations, metadata, links, image organization, and inbox review.

_{Collapsible note workbench}

Select text in the editor and the actions appear beside the passage. Translation language is available directly in the toolbar, while the remaining actions stay one click away.

_{Translate, explain, polish, expand, summarize, annotate, or ask}

Notes attached to the passage, not inserted into it

Annotations are stored in plugin data instead of being appended to the Markdown body. Write your own thought or save a translation, then return to the marked passage later. Annotations can be edited, deleted, located, and exported.

_{Write a thought, question, or TODO}

_{A lightweight anchor marks the source passage}

Models are profiles, not global credentials

Each profile carries its own endpoint and key. Keep several text or vision models, select one active text model and one active vision model, and route translation to a smaller dedicated text model when desired.

Images are part of the note context

Paste or drag images into chat, or let AI Organizer discover images referenced by the current note or selection. Limits for image count and file size are configurable per request; they do not limit how many images a document may contain.

Formatting stays reviewable

Formatting runs in a non-blocking workflow and opens a dedicated review window when the model finishes. Compare the original, formatted result, and diff before applying the change.

Installation

Obsidian Community Plugins

1. Open Settings -> Community plugins in Obsidian.
2. Select Browse and search for AI Organizer.
3. Select Install, then Enable.

You can also open the AI Organizer plugin page directly.

BRAT

Install BRAT, choose Add a beta plugin for testing, and enter:

Pitkil/obsidian-AI-organization-plugin

Manual installation

Download main.js, manifest.json, and styles.css from the latest release, then place them in:

<your-vault>/.obsidian/plugins/ai-organizer/

Reload Obsidian and enable AI Organizer under Community plugins.

Configuration

Open Settings -> AI Organizer. Configure at least one usable text-model profile before using chat or AI tools.

Model profiles

OpenAI-compatible profiles work with OpenAI, DeepSeek, Qwen, GLM, Kimi, Ollama, LM Studio, vLLM, and other services that expose a compatible chat-completions API. Dedicated adapters are included for Anthropic Claude and Google Gemini.

Text, vision, and OCR

• The active text model handles chat, writing, translation, formatting, metadata, inbox classification, and link suggestions.
• The active vision model handles note images and chat image attachments. Only the selected vision profile is used for a request.
• Quick profile presets are available for local Ollama and LM Studio vision endpoints; no vision-model weights are bundled with the plugin.
• When no vision model is configured or its request fails, the built-in tesseract.js path extracts visible text and passes it to the text model.
• OCR is a text-extraction fallback, not a replacement for visual reasoning. Formulae, layouts, charts, and photographs are better handled by a vision model.

Prompt customization

AI Organizer exposes two practical prompt layers:

1. Chat system prompt controls the assistant's default role, language, tone, and response style.
2. Formatting Prompt templates let you save named formatting instructions, edit or remove them, and select them as the default formatting mode.

Built-in formatting modes remain available for full formatting, Markdown syntax, heading structure, and CJK/English spacing and punctuation.

Translation

Set a default target language, maintain a list of frequently used target languages, and optionally choose a dedicated fast or inexpensive text model. If no translation model is selected, the active text model is used.

Feature Reference

Selection workflow

Applied text edits are briefly highlighted and expose an undo action. Obsidian's normal undo history remains available.

Chat and context

• Toggle current-note and selection context independently.
• Paste or drag image attachments into the composer.
• Stream model responses and stop generation.
• Switch between configured, usable text-model profiles below the input.
• Estimate context use from the note or selection, history, current input, images, and the model's configured context window.
• Restore recent conversation history after reopening the sidebar.
• Save a conversation as Markdown or clear current history.
• Collapse the workbench to preserve conversation space.

Formatting and organization

Commands

Privacy and Limits

• AI actions run only when you trigger them. Depending on the operation, the plugin may read the active note, selected text, referenced images, candidate note titles, or attachment paths.
• Content sent to a remote provider is governed by that provider's terms. Local Ollama or LM Studio endpoints can keep model requests on your machine or local network.
• API keys are stored in Obsidian's plugin data. They are not encrypted by AI Organizer; protect and sync that data accordingly.
• OCR uses tesseract.js. Required runtime or language data may need to be obtained and cached before local recognition can run.
• Long note content is truncated before requests. Image limits are configurable from 1–200 images and 1–50 MB per image; defaults are 20 images and 5 MB.
• The context ring is an estimate, not provider billing data. It uses the configured context-window size and approximate token counts.
• Annotation records live in plugin data. Sync plugin data if you expect the same annotations on multiple devices.
• Review generated edits, metadata, links, and file-move suggestions before applying them to important vaults.

Development

npm install
npm run dev
npm run build
npm test

src/
├── main.ts          Plugin lifecycle, commands, editor interactions
├── settings.ts      Persisted settings, defaults, migration
├── providers/       OpenAI-compatible, Claude, Gemini
├── core/            Chat, formatting, OCR, images, metadata, inbox, links
├── ui/              Sidebar, settings, review and workflow modals
└── utils/           Position, Markdown, notification helpers

Contributing

Issues and focused pull requests are welcome. Please describe the user workflow being changed, keep unrelated refactors separate, and run npm run build plus npm test before submitting.

License

MIT © Wang Yilai