Help center › Bots & channels

Publish your knowledge as a bot.

In Keldura, a bot is a reusable AI chatbot defined once — its knowledge (Library collections), its instructions (prompts), and its model. A channel is where that bot talks to people: Telegram, a chat widget on your website, or AI agents via MCP. Define the bot once, connect it to any channel.

Step 1 — Define a bot

On the Bots page, create a bot and give it:

Knowledge — one or more Library collections it answers from. The bot only knows what's in those collections, which keeps answers grounded and citable.
Instructions — a prompt that sets its tone, scope, and rules (for example: "answer only from the docs; if unsure, point to the contact page").
Model — the AI model that powers it.

Update the collections and every connected channel picks up the new knowledge — no re-deploying.

Step 2 — Connect channels

Connect a Telegram bot on the Telegram page and link it to your Keldura bot. People chat with it in Telegram and get cited answers from your knowledge. Useful for communities, internal teams, and support groups.

Website widget

The Website widget page embeds a chat over your collections on any site you control (requires a paid subscription):

Create a widget and pick the collection or bot behind it.
Add the domains allowed to embed it — the widget refuses to load anywhere else.
Copy the embed snippet — a one-line script tag or an iframe — into your site.

Visitors can then ask questions right on your page and get answers grounded in your content.

MCP and the developer API

Expose knowledge to AI agents and your own software. The MCP page connects Keldura folders to MCP-capable AI agents (ask questions, fetch documents, list citations programmatically). The developer API offers the same over REST with API keys.

Good practices

Curate the knowledge — a bot is only as good as its collections. Import your docs, FAQs, and guides; remove stale material.
Scope with instructions — tell the bot what it should not answer and where to send people instead.
Test before publishing — chat with the collection yourself first; the answers the bot gives are the answers you see.
Keep it living — connect monitored sources so the bot's knowledge updates as your content does.