Turn Any Vault into a Public AI Agent

Most knowledge bases are private by design. You capture, organize, search — and the information stays with you. That works until you need to share it with someone who isn't in the system: a client, a visitor to your site, a colleague who doesn't have an account. At that point you're back to copy-pasting, writing summaries, or just telling people to email you.

Zibri's AI Agents feature flips that. You can turn any vault into a public-facing chatbot in a few clicks, publish it with a shareable link, and let anyone ask it questions — without giving them access to your account, without any coding, and without worrying that it will make things up.

One vault becomes one agent — and it only knows what's in that vault

The scope constraint is the key design decision. When you create an agent, you pick a single vault as its knowledge source. The agent answers only from the content in that vault — your notes, uploaded documents, transcripts, anything you've captured there. It pulls nothing from the open internet and nothing from your other vaults.

That matters for two reasons. First, it keeps the agent accurate: every answer is grounded in content you wrote or approved, the same Retrieval-Augmented Generation (RAG) engine Zibri uses for your private Chat. Second, it keeps the agent in scope. A support agent for one product line won't drift into answering questions about another. A research vault you've shared with a client won't leak notes from a different project.

Setting one up takes about two minutes

Go to Shared Agents in the sidebar and click New Agent. Select the vault you want to use, give the agent a name, and optionally write a system prompt to set its tone — formal, conversational, bullet-point responses, whatever fits the context. Publish it. You get a shareable link in the format /s/agent/abc123.

Anyone with that link can open the agent page without logging in to Zibri. They type a question, the agent retrieves the most relevant content from the vault, and it returns a sourced answer. Visitors see which notes or documents the answer drew from, so they can ask for the original if they want it.

No credentials required on their end. No setup. Just the link.

The analytics tell you what people actually want to know

The Shared Agents screen shows query counts and the questions each agent has received. That's more useful than it sounds. If you've published a vault of onboarding materials and the agent keeps getting the same question, that question isn't answered well enough in the vault — or isn't answered at all. The logs turn an invisible gap into a visible one.

The fix is straightforward: add a note or document to the vault that addresses it, and the agent picks it up immediately. No retraining, no configuration change. The vault is the model.

Where this is actually useful

A few concrete situations where an AI agent does work that a static FAQ page or a shared document can't:

Client-facing knowledge bases. You've done a research project or built out a resource library for a client. Instead of sending them a folder of PDFs, you give them an agent link. They ask questions in plain language and get specific answers with citations, rather than skimming forty pages hoping the right paragraph appears.

Internal documentation that gets asked about constantly. Policies, process guides, reference material. Build the vault, publish the agent, share the link with the team. Colleagues get answers faster than they would searching the document, and you stop fielding the same questions.

Product or service information. Put your product specs, FAQs, and case studies in a vault. The agent handles first-level questions from anyone with the link. Because it only answers from that vault, it won't speculate about things you haven't documented.

Personal expertise, made accessible. A consultant or researcher might have years of notes on a narrow topic. An agent built on that vault lets someone else query that expertise without the expert having to be present.

What it won't do — and why that's a feature

The agent will not answer questions outside its vault. If a visitor asks something the vault doesn't cover, the agent says so rather than guessing. That's the RAG architecture enforcing a constraint you'd want anyway: an agent that admits it doesn't know is more trustworthy than one that confidently fills gaps with plausible-sounding fiction.

Zibri's documentation is explicit that agents never draw from outside sources. The vault is the ceiling, not just the starting point.

The takeaway

AI Agents aren't a publishing feature bolted onto a notes app. They're what happens when you've already done the work of capturing and organizing knowledge in vaults, and you want that knowledge to do something — answer a client, support a team, surface what you know to someone who needs it. The vault you've already built is the hard part. The agent is just the door.

If you have a vault with content people ask you about, it's worth publishing an agent and watching what questions come in. The logs will tell you more about your gaps than another round of editing your own notes will.