Your Vault as a Chatbot: Zibri AI Agents

One vault, one agent, one source of truth

When you publish an AI Agent in Zibri, you pick a single vault as its knowledge base. The agent answers questions using only the notes, documents, and transcripts in that vault — and because Zibri's underlying retrieval is source-cited, every answer traces back to the exact note it came from.

The practical effect: if you haven't put something in the vault, the agent won't invent it. That constraint is the feature.

A straightforward use case is a product FAQ. Load your latest internal documentation into a "Product FAQ" vault, publish it as an agent, and hand the link to your support team or post it on your site. Customers ask questions; the agent answers from what you've written. When you update the docs, the agent's answers update automatically — because the vault updated.

What you can tune

The agent isn't just a raw query interface. When you create one, you can set a custom system prompt that shapes how it responds: formal or conversational tone, whether it replies in bullet points, whether it identifies itself by a name. A legal team might set the agent to respond conservatively and always flag that it's a summary, not legal advice. A sales team might configure a friendlier tone for a partner portal.

Zibri also gives you analytics on query volume — so you can see what people are actually asking, which is often more useful than any survey. If a particular question keeps appearing that the agent can't answer well, that's a signal your vault needs a new note.

Sharing without giving away the keys

There's a meaningful difference between sharing a vault's contents and sharing access to the vault itself. AI Agents keep that line clear.

The person querying your agent doesn't get to browse your notes, edit anything, or see vault structure. They get a chat interface that answers from your content. You stay in control of what goes in the vault; they get the output.

This is different from Zibri's password-protected note sharing, which lets you hand someone a single note via a secure link. Agents are for when you want to make a whole body of knowledge queryable — without handing over the raw files.

The use cases that make this concrete

Internal knowledge base. An operations team keeps runbooks, process docs, and decision logs in an "Ops" vault. Instead of answering the same onboarding questions repeatedly, they publish an agent so new hires can query it directly. The agent only knows what the ops team has documented — which creates healthy pressure to keep that documentation current.

Customer-facing FAQ. A company publishes a "Product FAQ" agent on their support page. Answers are always drawn from the latest internal docs. When pricing changes, update the note; the agent's answer changes with it.

Partner portal. A sales lead gives a strategic partner access to a "Partner Insights" vault through an agent link. The partner can ask questions and get substantive answers — without the sales lead needing to be on every call.

Research synthesis. A consultant running a long engagement loads interview notes, reports, and client documents into a project vault, then uses the agent as a query interface when preparing deliverables. The sourcing is automatic; the agent cites which note each piece of information came from.

What the agent cannot do — and why that matters

The agent will not answer questions outside its vault. It will not pull in web results, guess at information that isn't there, or stitch together answers from other vaults you own. If a customer asks something your documentation doesn't cover, the agent says so.

This sounds like a limitation. It isn't. It means you can tell users — and mean it — that every answer came from content you reviewed and approved. For regulated industries, compliance use cases, or anywhere accuracy is load-bearing, that constraint is the entire point.

[PICTURE: screenshot of the Zibri AI Agent interface showing a vault selected, a custom system prompt field, and a sample query with a source-cited answer]

Getting started

Creating an agent takes [FIGURE NEEDED: number of steps or time to publish an agent]. Pick a vault, write a system prompt if you want one (you can leave it blank), and publish. Zibri generates a shareable link. Anyone with the link can query it; they don't need a Zibri account.

The harder work is the vault itself — making sure the notes are complete, current, and clearly written. That's not a Zibri problem; it's the same work you'd do to keep any documentation useful. The agent just makes that documentation instantly queryable by anyone you want to reach.

If your organization has ever said "we have all this knowledge, we just can't get people to it fast enough," an AI Agent is the most direct answer Zibri offers. Build the vault once, publish the agent, and let the content do the work.

— Erin