Obsidian Personal Assistant

An Obsidian plugin which help you to automatically manage Obsidian.
中文手册 · Manual

💯Tips: If you are not a developer, please refer to the manual for optimal use.

NOTE: Pagelet discovers connections across permitted notes and lets you inspect their sources or continue in Chat. See the current Pagelet guide. Also supports LLM chat with Memory. When Memory is enabled and an AI provider is configured, the first Chat can prepare Memory in the background without a blocking prompt: eligible note text is sent to the configured embedding provider and may use API credits. You can turn Memory off in Settings; recovery, settings-change, and manual rebuilds still ask before costly work.

Historical v2.7 guide: The archived v2.7 user guide and Chinese version describe that release's workflows and video script. Use the Manual and current guides for repository behavior. Documentation was reconciled with the repository on 2026-10-06; available features in an installed build depend on its actual version and release record.

Project docs: Use the documentation index for repo-local requirements, decisions, product/architecture contracts, active development, backlog, and historical records.

Personal Assistant demos

macOS demo: AI Chat with Memory, note understanding, visual relationships, and Pagelet review workflows.

Personal-assistant-macOS-demo-1080p-public.mp4

iOS demo: Mobile Chat, Mermaid rendering, featured images, and connection discovery.

Personal-assistant-iOS-demo-1080p.mp4

AI Chat in 60 seconds

Three steps to get talking to your notes:

1. Click the Personal Assistant icon on the left ribbon — it opens the AI Chat view directly. (Right-click the same icon for the older plugin controls modal.)
2. If AI is not configured, the Chat banner lets you choose Qwen China, Qwen International, or OpenAI and add a token without leaving Chat. If only the token is missing, it preserves your existing provider settings and asks only for the token.
3. Select Start to save and continue. Use Advanced setup for a custom endpoint or model; the first Settings visit keeps AI Provider open and the other groups folded so the required fields stay in focus.

See the AI Chat chapter in the Manual for prompts, citations, web search, and Memory tips.

RAG-Search.mp4

---

Supporting featured image generation by AI according to the content of the note.

featured-images-ai-generation.mp4

---

AI Helper to improve your Obsidian notes management

Animation rendering statistics

Preview records

List callout

Update metadata

Update Plugins and Themes

Basic Usage

Features

NOTE: The currently supported features are all from my personal needs, feature request is welcome by submitting issues.

1. automatically create note in the specified directory with the configured file name
2. automatically open current note related graph view
3. automatically open Memos like quick note in macOS
4. open Personal Assistant controls from the command palette
5. automatically update plugins with one command
6. automatically update themes with one command
7. automatically set color of graph view
8. list all callouts css configuration for quickly inserting
9. chat with AI using Memory from your notes, or answer immediately without reading memory
10. Pagelet — discover connections from the current note, inspect source-backed insights, and continue the discussion in Chat; discovery itself does not modify notes

Develop

Please reference HERE.

Memory preparation performance note

Since 1.6.4, rebuilding Memory batches note chunks across files and uses provider-aware embedding limits instead of a fixed per-file delay. Qwen text-embedding-v4 / text-embedding-v3 rebuilds send up to 10 chunks per request with token-aware throttling and retry feedback. The long-running Memory notice now reports live progress such as scanning notes, embedding chunks, writing the index, retrying, and ready.

Manual "Update memory" keeps the safer per-file refresh path for now, but it also reports file-level progress and still skips unchanged notes before calling the embedding provider. Sharing the global rebuild batching pipeline with refresh is planned as a later large-vault optimization.

Background memory maintenance note

After the first background preparation or an approved recovery/manual preparation succeeds on a device, changed notes can be maintained automatically while Obsidian is open. Chat no longer waits for a refresh when the local SQLite/WASM Memory index is ready; it can answer with the last prepared Memory while a background reconcile/refresh updates changed notes.

Automatic maintenance writes Memory embedding data to the device-local SQLite/WASM OPFS backend and keeps VSS maintenance state in local Obsidian app storage. It does not create new vss-index-state/, vss-index-state/<deviceId>/manifest.json, or vss-cache/dirty.json files in the vault.

Network and privacy note

Personal Assistant does not upload telemetry or analytics. An optional, default-off capability-usage setting emits local diagnostic events containing only capability/provider IDs, status, and duration; it excludes prompts, note text, paths, URLs, credentials, and model output. See the local usage-event contract. By default, Statistics history is stored in local Obsidian app storage on the current device and is not uploaded by the plugin. If you enable cross-device Statistics history, the plugin creates vault-visible Statistics history files so your normal vault sync can carry them; Git users will see those files change.

Feature	Trigger	Data sent	Destination	Background?	User control
Chat	You send a message	Prompt; selected note/tool context, Memory search query and excerpts when enabled; processed copies of attached images when the Chat model supports images	Configured AI provider	No	Provider, chat, Memory, and attachment controls
AI note tools	You run summary or note AI actions	Current note content and the generated prompt	Configured AI provider	No	User action and AI settings
Memory prepare/update	First Chat while Memory is enabled and the AI provider is configured; or an approved recovery/manual action	Eligible note text and Memory search data	Configured embedding provider	First-use preparation runs in the background; manual actions block on progress; after success, changed notes may update in background	Memory on/off, Data Boundary exclusions, provider settings, and background toggle
Memory changed-note maintenance	Memory has been prepared and background updates are enabled	Changed note text	Configured AI provider	Yes	Memory background setting
Qwen web search	You enable web search for Qwen responses	Question and final prompt context	DashScope/Bailian	No	Qwen response setting
Featured image generation	You run image generation	Current note content for prompt generation, then image prompt and task requests	Configured AI provider and DashScope/Bailian	Polls task status after your request	User action and AI settings
Chat image generation/editing	You explicitly request an image or edit	Image description and authorized reference-image copies; selected note text when used to prepare the description; GET requests to retrieve completed images	Configured Chat/image connections; service-provided Alibaba Cloud OSS image-result URLs	Task status may be polled after submission; completed images are downloaded for local saving	Explicit request, source selection, and image connection settings
Share Card remote images	You open Share Card for content containing remote image references	GET requests to the referenced image URLs; the plugin does not add the note body or AI credentials to these requests	The image hosts referenced by the content, including supported image references within SVG	Resource preparation starts when the Share Card opens	User action and the content's image references
Pagelet discovery	You run discovery or enable its automatic preparation	Permitted anchor/source note text and optional web queries	Configured AI provider; configured supported web search	Automatic preparation may run in background	Pagelet settings and Data Boundary exclusions
Ghost preparation/update	You explicitly use @blog2ghost	Selected article and required media to Ghost; note text used for metadata preparation to the configured AI provider	Configured Ghost site and AI provider	No	Explicit workflow, desktop-local key, first publication in Ghost, and exact-candidate update confirmation
Plugin/theme updater	You run the updater/install flow	Plugin or theme IDs and download requests	GitHub and jsDelivr	No	User action

See image chat and storage and the Ghost workflow for their data, storage, and publication boundaries.

Image hosts receive the requested URL, including any query parameters already present in the image reference, and normal request metadata.

The bundled LangChain dependency contains an optional tokenizer-dictionary loader for https://tiktoken.pages.dev/js/<encoding>.json. Personal Assistant currently estimates prompt sizes locally and does not invoke that loader in its Chat call paths. If a dependency token-counting path invokes it, it downloads the public encoding dictionary without sending prompts, note text, or AI credentials; the server still receives normal request metadata. A domain appearing in the bundle identifies a potential dependency request, not an observed request on every Chat interaction.

Base64 usage note

Personal Assistant and its bundled dependencies use Base64 for binary transport and local asset loading: encoding processed image attachments and reference images as data URLs, reading supported inline raster images in SVG and Share Card resources, carrying bundled fonts as data URLs, and decoding the SQLite WASM binary. A local worker data-URL compatibility path also decodes Base64. Worker code and the WASM module are executable packaged assets; the SQLite integration and its dependency are described below.

Memory pagination also uses Base64url to carry JSON query state, a state fingerprint, and an offset between tool calls. These cursors may contain query filters; Base64 is reversible encoding, not encryption or a credential-protection mechanism. Dependency utilities also contain Base64 conversion/validation paths, such as binary embedding-response decoding.

VSS SQLite/WASM dependency note

The local Memory index uses sqlite3.wasm from the official @sqlite.org/sqlite-wasm package, pinned to 3.53.0-build1. SQLite stores local Memory metadata, note chunks, and embeddings. The build embeds the package's WASM binary and JavaScript loader together with PA's SQLite worker code in main.js; when needed, the plugin creates local Blob URLs for the worker source and decoded WASM bytes, then starts the worker. It does not download the WASM module from an external server. Preparing or updating Memory can still send note text to the configured AI provider, as disclosed above.

The WASM module imports nine functions from wasi_snapshot_preview1: clock_time_get, environ_get, environ_sizes_get, fd_close, fd_fdstat_get, fd_read, fd_seek, fd_sync, and fd_write. These provide clock, environment, and file-descriptor support through the bundled Emscripten JavaScript implementations. In PA's worker integration, file descriptors refer to the Emscripten virtual filesystem and streams; environment values are constructed by the runtime. They are not bindings to the host operating system's filesystem, environment variables, or process-launch APIs. SQLite database persistence uses a separate browser OPFS VFS in the worker. This module has no socket or process-launch imports. The WASI import names alone do not establish broader host access or a malware verdict.

Raw WASM fingerprint (verified on 2026-10-07):

Field	Value
Package	@sqlite.org/sqlite-wasm@3.53.0-build1
File inside the package	dist/sqlite3.wasm
Raw byte length	864752
SHA-256 of the raw WASM bytes	02d7e48164395fa68f81c6ec33e9da5461be397dc57602ac0cd89b4bbba1d312

The cached npm archive was checked against the SHA-512 integrity recorded in package-lock.json. Its WASM file, the installed dependency file, and the sole WASM module decoded from the local dist/main.js were byte-for-byte identical and had the fingerprint above. This verifies package-to-bundle byte consistency; it does not prove a reproducible build from upstream C source or identify a Community scan's module without that scan's artifact fingerprint. Recheck the fingerprint when the dependency or binary changes.

See the SQLite WASM documentation, Emscripten filesystem overview, and third-party notices for provenance and licensing. Before publishing a release with this backend, review the upstream package license and release terms for your distribution scenario.

License and commercial boundary

Starting with version 2.8.0, the Personal Assistant client source is licensed under AGPL-3.0-only. Historical releases are not relicensed retroactively; previous tags and artifacts retain the license notices and metadata published with those releases. For the current release, use the exact GitHub tag, source archive, LICENSE, NOTICE, and THIRD_PARTY_NOTICES.md files as the source and legal reference.

Version 2.8.0 is a license and compliance migration release. It does not introduce an account system, license key, checkout flow, feature lock, hosted commercial service, or paid entitlement check. Future hosted services, support, warranty, privacy, or trademark terms may be separate, but they must not restrict AGPL rights to use, modify, and redistribute the client.

The Personal Assistant name, personal-assistant plugin ID, icons, logos, and marketplace identity are not licensed to forks by the AGPL client license. See TRADEMARKS.md, NOTICE, and THIRD_PARTY_NOTICES.md.

Mobile VSS validation note

The local VSS SQLite/WASM backend has been smoke-tested on Obsidian Desktop and Obsidian iOS with the test vault, including rebuild, refresh, reload persistence, chat, and Memory references. Android has not been fully validated on a physical device yet because no Android test device is currently available, so Android VSS support should be treated as pending verification.

Install

Now Personal Assistant plugin is available in plugin market, you can install this plugin directly within Obsidian App, please check this mannual to get more details.

Install

• Download from the release

Install with BRAT

• Install BRAT from the Community Plugins in Obsidian
• Open the command palette and run the command BRAT: Add a beta plugin for testing
• Copy https://github.com/edonyzpc/personal-assistant into the modal that opens up
• Track the latest release, or freeze a specific beta tag if you are testing a targeted build
• Click on Add Plugin -- wait a few seconds and BRAT will tell you what is going on
• After BRAT confirms the installation, in Settings go to the **Community plugins ** tab.
• Refresh the list of plugins
• Find the beta plugin you just installed and Enable it.

Maintainers should use the repository BRAT release workflow before inviting beta testers: docs/operations/brat-beta-testing.md.

Manually Install

• Build with commandline: npm install && npm run build or download from release page
• Copy over main.js, styles.css, and manifest.json to your vault config folder, usually {VaultFolder}/.obsidian/plugins/personal-assistant/. If your vault uses a custom config folder, use that folder instead of .obsidian.

Use

1. Create note in specificed directory

• Open the command palette and find the command
• New note is created and start your recording
• [Recommendation] Use Folder Templates of plugin Templater to format the created notes by the command above, the example is as following

2. Open memos in hover editor

• Open the command palette and find the command
• Do anything you like in memos

3. Open graph view of current note

• Open the command palette and find the command
• Open setting tab for more customize
• Navigate your current note graph view with backlink and outgoing link
• configure color of graph view

4. Enable/Disable plugins for obsidian with one command

• Open the command palette and find the command
• Select the suggestion to enable/disable plugin(or you can search the plugin by its name)
• [Note] In suggestion tab, the green checkbox means plugin is already enabled and the red uncheckbox means plugin is already disabled

5. Update plugins for obsidian with one command

• Open the command palette and find the command
• Trigger the command to update plugins
• See the updating result which is displayed in the right corner

Attribution

• Best thanks for project obsidian-advanced-new-file for the code of createNote, createDirectory
• Best thanks for project obsidian-callout-manager for the class CalloutPreviewComponent and color.ts
• Best thanks for project better-word-count for the package stats

Contact

If you've got any kind of feedback or questions, feel free to reach out via GitHub issues.