Private AI search for your team, on your infrastructure.
The same cited answers as the personal app, run as one container you host. Your index and your documents stay on your servers. Reach out for a quote.
Personal vs Team
| Feature | Personal | Team · Self-hosted |
|---|---|---|
| Where it runs | The desktop app on your computer (macOS, Windows, Linux) | One Docker container on your infrastructure |
| Install | One download, no Docker | docker compose up from the repo, or the prebuilt image |
| Where data lives | A database on your disk | A SQLite file on the container's /data volume — no database server |
| Models | Local Ollama, or your own OpenAI-compatible API | Same |
| Sources | Folders, Drive, Jira, Linear, Miro, JSON APIs | Same |
| Sign-in | None — it only listens on your machine | Optional password sign-in; required before it will start on a non-local address |
| License | Free for personal use | Required for team and commercial use; checked offline at startup |
What you can check
Each of these is in the public source, not on a roadmap. Network calls, secrets storage and known limits are on the security notes; coming from NotebookLM? See the comparison.
Source labels gate cloud AI
Label a source internal, public or sensitive. Sensitive sources are kept out of answers from a cloud model you have not approved.
Refusals leave a record
When the label gate excludes sources, the app writes an event to its system log. If nothing survives the gate, it refuses to answer instead of guessing. Every agent ask and search call is logged too.
Public-only mode for agents
The MCP server has a public_only switch: answers come only from sources labelled public, so a coding agent can be pointed at what is safe to share.
Portable backups
Export the whole index to one file and import it elsewhere, with a dry-run preview first. Records are matched by stable IDs, not local row numbers. Credentials stay behind unless you opt into a passphrase-encrypted bundle.
Cited answers
Every answer lists the document and section it came from, in the browser and for coding agents.
Your model choice
Run models locally with Ollama, or bring your own OpenAI-compatible API. Without any model, keyword search still works.
Deployment
One container, one process, state on one volume. Copy the example environment file, add your license and model settings, then:
A prebuilt image is published with each release. Put TLS in front with your usual proxy. Full steps: hosting/README.
Request a quote
Tell us a little about your team and we will reply by email.
We use these details only to reply to your request. Handled by Formspree as our processor — see the Privacy notice.