An AI Assistant You Can Actually Trust

Most AI assistants ask you for trust up front. They act, and you find out afterward. That is a bad fit for a tool that holds your knowledge and can change it.

Arivu’s assistant works the other way around. It can suggest and draft, but it cannot change anything on its own. Every action it wants to take becomes a reviewable proposal that sits still until you approve it. The default is safety, and you are always the one who says go.

Reference visual generated with an image model, not a product screenshot.

Propose, Review, Approve

The assistant follows one strict rule: it never mutates your second brain directly. Instead, it moves through a ledger you control.

You can ask the assistant for suggestions from your Inbox, your Review queue, a search, or a specific saved item. It returns inert drafts: proposals for a small set of allowlisted actions, such as updating an item’s stage, creating a link between two items, adding a task, or setting a reminder. A draft is just a description of what would happen. It does nothing yet.

From there, the proposal waits in a ledger. You can approve it, and only then does Arivu carry it out. You can reject it, and it disappears with nothing changed. If an approved action fails to run, that is tracked too, so the ledger stays honest about what actually happened.

What this means for you: You get the leverage of an assistant that reads across your whole collection, without ever handing over the keys. The assistant proposes. You dispose.

Answers From Your Own Library

Search can now do more than list matching pages. When you ask a question, Arivu can synthesize a cited answer drawn only from your own saved content: your summaries, highlights, snippets, linked context, and notes. Every claim points back to the items it came from, so you can follow the citation and check the source.

This is deliberately grounded. The answer is not the open internet’s opinion. It is what you saved, pulled together and cited, so you can trust where it came from.

What this means for you: Ask your library a question and get a straight, sourced answer built from material you already chose to trust.

Useful Even With No AI Key

Here is the part that matters most for self-hosters: Arivu does not fall over without an AI provider.

When no provider key is configured, Arivu still processes what you save using local, deterministic enrichment. You still get summaries, key bullet points, highlighted quotes, suggested tags, and the entities and concepts that feed the knowledge graph. It is not the same as a large model’s output, but it is real, useful structure, generated on your own machine with nothing sent anywhere.

Connect a provider like Gemini and the AI-powered version takes over. Leave it disconnected and the app still works end to end.

What this means for you: AI is an upgrade, not a dependency. You can run a completely self-contained Arivu that never calls an outside service, and it still organizes your knowledge.

Other Changes

• Suggestions are scoped to context. Ask from the Inbox, Review, a search, or a single item, and the assistant works from that context instead of guessing.
• Allowlisted actions only. The assistant can propose item state changes, links, reminders, and tasks. It cannot invent new kinds of changes.
• Plain language in the UI. Proposals talk in terms of what is pending and what will run, not internal ledger jargon.

An assistant that holds your knowledge should earn trust by design, not ask for it by default. Arivu’s assistant proposes and cites, runs only when you approve, and steps aside entirely if you never plug in an AI key at all. Powerful when you want it, quiet when you do not, and yours the whole way through.