AI agents built around your data privacy

Build agents without code, ground them in your documents, and choose where the model runs: Switzerland, the EU or global.

Start freeRead the docs

Everything an agent needs, set up without code

Your choice of where the model runs

Pick a model region in Switzerland, the EU or global, bring your own API keys, or deploy on-premises for full isolation.

Knowledge bases with retrieval testing

Group text, PDF and Word files, crawled links and Q&A into folders you reuse across agents, then test retrieval before you publish.

Your APIs and MCP servers as tools

Add REST APIs with their auth and schemas, or MCP servers over SSE or Streamable HTTP, and debug each call.

One chat for every model

Your team talks to its agents in one chat, whichever model runs them, and can switch models without rebuilding the agent.

Guardrails

Redact personal data, detect prompt injection, set risk thresholds, and block the words, patterns and topics you rule out.

Publish where people work

Embed an agent in your site with one script tag, share it as a standalone page or QR code, or call it from a Logic workflow.

Composability for your AI use cases

Three foundation layers. One application layer. Focus on the application, not the orchestration.

  1. ModelsConnect multiple models from any provider, or self-host your own. Switch without rebuilding.Model Foundation
  2. Knowledge BasesCreate custom knowledge bases for RAG-powered agent responses.Agent Foundation
  3. Tools & MCPConnect your APIs and MCP servers for agentic AI workflows.Agent Foundation
  4. WorkflowsVisual workflow builder. JSON DAG under the hood. No custom tooling needed.Workflow Foundation

Application Layer

  • Modular
  • Interoperable
  • Agile

Focus on the Application without worrying about AI Orchestration

From a template to a published agent

Agents are built in the console at agents.seliseblocks.com. Your site talks to them through the embed script, and Logic workflows call them with the AI Agent node.

  1. Start from a template

    Pick customer service, IT support or HR assistant, or start blank. Sequential agents answer in one pass, and reasoning agents plan answers over several steps.

  2. Add knowledge, tools and guardrails

    Connect knowledge base folders, add your APIs and MCP servers, choose the model, and set the guardrails.

  3. Test and publish

    Try the agent in the playground, read each answer's steps, tools and sources in the conversation log, then publish it.

<script
  src="https://gpt.seliseblocks.com/embed.js"
  data-widget-id="<widget-id>"
  data-widget-type="chat"
  data-project-key="<x-blocks-key>"
  data-app-domain="https://<your-app-domain>"
  data-app-mode="prod">
</script>

Models and tools

  • OpenAI
  • Anthropic
  • Azure OpenAI
  • OpenRouter
  • Your own model endpoint
  • MCPSSE
  • MCPStreamable HTTP

Recent changes to Agents

All Agents updates

Questions about Agents

Where do my agents' models run?

You choose the model region: Switzerland, the EU or global. You can also bring your own API keys, connect your own model endpoint, or deploy on-premises.

Which models can an agent use?

Preconfigured Blocks Models, which are OpenAI models hosted on Azure, the OpenAI and Anthropic APIs with your own key, deployments on Azure or OpenRouter, and custom or self-hosted models through an API URL.

How do I add an agent to my site or app?

Publish it and paste the embed script into your site, share it as a standalone page, or call it from a Logic workflow with the AI Agent node.

What can an agent use as knowledge?

Text and Markdown, files such as PDF and Word documents, website links it can crawl again periodically, and question-and-answer pairs, kept in folders you can reuse across agents.

Start with a free project

Sign up at os.seliseblocks.com and create a project, or paste one prompt into Claude Code, Codex or Cursor and let your agent set it up through the Blocks CLI.

Start freeSpeak to a Consultant