Note Companion: An Obsidian Plugin That Turns Captures Into Notes, With a Self-Hosted Backend

What Note Companion Actually Does

Note Companion is an Obsidian plugin that processes raw captures: audio files, YouTube links, meeting recordings, and loose notes. It transcribes audio and video, generates summaries and structured notes, and suggests folders, tags, titles, and templates. It also includes a chat interface that can reference specific files, folders, or tags via @mentions. The README positions it for researchers, students, and anyone with an Inbox folder full of unprocessed notes. The key distinction from a generic AI chat is that it operates on your vault's context, not just a single prompt. This is not a note-taking app; it is a processing layer that sits inside Obsidian.

The Architecture: Cloud, Self-Hosted, or Local

The plugin itself is a thin client. All AI processing happens on a backend. There are three modes. The default is Note Companion Cloud: you create an account, paste a license key into the plugin, and your files are sent to Note Companion's API. The second is self-hosting: you deploy the backend yourself (via Docker or pnpm build:self-host), set your own provider keys in the server's .env file, and point the plugin to your server URL. The third is local Ollama, which is experimental and chat-only. The README is explicit that the plugin has no screen for entering OpenAI, Claude, or Gemini keys directly; those keys live on the backend. This separation is a deliberate design choice. It keeps the plugin simple, but it means that even self-hosted users must run a separate service. If you are not comfortable running a Node.js backend, the cloud path is the only practical option.

Installation and Setup: Real Commands and Config Keys

Installation is standard for Obsidian: Settings, Community plugins, Browse, search for Note Companion, then Install and Enable. The plugin is marked desktop-only in its manifest, so it will not load on iOS or Android. After enabling, you go to Settings, Note Companion. For the cloud path, you create an account at notecompanion.ai, copy a license key from the dashboard's API Keys section, and paste it into the General tab's License Key field. For self-hosting, you deploy the backend following SELF-HOSTING.md. The README mentions Docker or pnpm build:self-host. You then add provider keys to the server .env, for example OPENAI_API_KEY, ANTHROPIC_API_KEY, MODEL_PROVIDER, MODEL_NAME, and optionally OPENAI_API_BASE for Ollama. In the plugin, you enable Advanced, Enable Self-Hosting, and set Server URL to something like http://localhost:3000. You can leave the License Key empty if ENABLE_USER_MANAGEMENT=false. A quick test is to open the Command Palette and run Note Companion: Open Chat.

The Meeting Workflow: Where It Shines

The README highlights the meeting recorder as the best demo. You open the Meetings tab, start recording, stop, select a note, and choose Enhance Note. The output includes embedded audio, a transcript, discussion points, and action items. This is a concrete workflow that combines several features: audio transcription, note formatting, and structured output. It is easy to see how this saves time for people who attend many meetings. The same mechanism handles YouTube links: paste a link into chat, and you get a transcript, summary, and optional structured notes. The README notes that YouTube transcription works even when captions are not available in the usual way, which implies some fallback logic, but the details are not specified. For large audio files, the README warns about a 25MB limit and suggests splitting or compressing. That is a real constraint to plan around.

Limitations and Failure Modes

The most obvious limitation is platform: desktop Obsidian only. The README states that the plugin will not load on mobile and that mobile users may see it as unsupported. There is a separate mobile app, but development on it is paused. That is a hard stop for anyone who primarily uses Obsidian on a tablet or phone. Another limitation is that the plugin has no built-in API key entry; if you want to use your own keys, you must run the backend. That adds operational overhead. The troubleshooting section lists common failures: transcription fails due to provider key, model availability, or file size; YouTube import fails if the video is private or restricted; local models fail due to CORS or network settings. The plugin also depends on a running backend; if the server is down, the plugin cannot connect. The README suggests checking the health endpoint with curl http://localhost:3000/api/health. There is no offline mode for the AI features. If you have no internet or no backend, the plugin is effectively useless.

Privacy and Data Handling

Privacy depends entirely on the mode you choose. In the cloud mode, files are processed through Note Companion's API, and usage statistics are collected for billing and rate limits. The README claims that vault content remains private and is only processed when you explicitly use features, but that is a claim, not a guarantee. In self-hosted mode, all AI processing runs on infrastructure you control, and there is no cloud usage tracking when ENABLE_USER_MANAGEMENT=false. This is the clear privacy advantage of self-hosting. However, self-hosting means your provider keys sit on your server, and you are responsible for securing that server. The README points to SELF-HOSTING.md for security notes, but does not summarize them. If you are handling sensitive research or client data, the self-hosted path is the only one that keeps data off third-party servers, but it requires you to trust your own infrastructure.

Maintenance and Upgrade Considerations

The project is under active development, with releases like 3.6.31 (Smarter Image Annotation & Error Handling) and 3.6.30 (Smarter YouTube Note Magic). The release cadence suggests frequent updates. For users, that means the plugin will change, and you may need to adapt to new settings or features. For self-hosters, you must also update the backend separately. The README does not describe an upgrade path, but it is reasonable to assume you need to pull the latest backend image or rebuild. The license is MIT, which is permissive: you can modify and redistribute the code, but you must include the original copyright notice. The project also has a cloud service that is not open source, so if you use the cloud path, you are relying on a proprietary service. If you self-host, you are running an open-source backend, but the plugin itself is also MIT. There are no obvious licensing traps, but you should check the SELF-HOSTING.md for any additional terms.

Alternatives and the Core Trade-Off

The main alternative is to use a general-purpose AI tool outside Obsidian, such as a standalone transcription service or a chat interface like ChatGPT, and then manually copy the results into your notes. That approach does not require a backend or a license, but it lacks the vault context and the automation. Another alternative is to use Obsidian's built-in features plus a separate automation tool like Templater or QuickAdd, but those do not provide AI transcription or chat. The core trade-off with Note Companion is convenience versus control. The cloud path is the easiest, but it sends your data to a third party. The self-hosted path gives you control but requires you to run a backend and manage provider keys. There is no middle ground where you can paste an API key into the plugin itself. If you are already comfortable with running a small server, self-hosting is a reasonable choice. If not, you must decide whether the cloud service's privacy policy is acceptable.

Editorial conclusion

Adopt Note Companion if you are a desktop Obsidian user with a heavy capture workflow (voice memos, YouTube links, meeting recordings) and you want AI-assisted organization without leaving the app. Avoid it if you rely on Obsidian mobile, need a fully offline solution, or expect to paste API keys directly into the plugin. Before adopting, verify your AI provider's file size limits, confirm your backend can reach the health endpoint, and test transcription on a small file. Choose self-hosting if you want to keep vault content off third-party servers; otherwise the cloud path is simpler but sends data to Note Companion's API.

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes