Embeddable AI agents
Blueprint indexes your files, PDFs, and web pages, connects the tools you need, and exposes a chat agent you can publish to any page — as a self-contained snippet that talks to your own API.
No agent to demo yet.
Create one in the builder and this panel goes live.
01 · Capabilities
Documents are split into clean paragraphs and indexed for retrieval. Every hit knows which source it came from, and answers are grounded in those passages.
Connect file, database, and third-party servers over MCP. Tools arrive namespaced by server, and you enable them per agent — no code changes required.
One self-contained snippet, no framework and no dependencies, served by open CORS-enabled endpoints. Paste it into any site, app, or admin panel.
02 · Live demo
The bubble below is the same self-contained snippet you'll copy from the builder's Deploy step — rendered here, shadow DOM and all.
No agent to demo yet.
Create one in the builder and this panel goes live.
03 · How retrieval works
Text, markdown, source code, PDFs, and web pages are read and normalized into clean paragraphs.
extract_paragraphs()
Paragraphs are embedded and organized into a summary tree, so related passages cluster and retrieve together.
RAPTOR indexer
The best passages are gathered with their source documents and passed to the model, so answers carry citations.
[1](doc A) · [2](doc B)
Create an agent, add your knowledge, and publish it anywhere.