Skip to content

The AI-native knowledge base for Open WebUI

Open WebUI is the self-hosted ChatGPT for the model you choose to run. Hjarni is the knowledge base it reads through Open WebUI's external tools.

Get started free See how it works

Free to start. No credit card required.

What this unlocks

Workflows that actually use your context

Long-term memory for your local model

Ollama and Llama-class models forget between chats. Hjarni gives them a persistent set of notes to search and read, the same way ChatGPT and Claude already work with it.

Keep private context off the cloud

If you run Open WebUI to avoid sending data to third-party AIs, Hjarni stays on the same side of that line. Your model reads notes from one knowledge base. You control the boundary.

Reuse the same notes across rooms and personas

Open WebUI lets you spin up rooms and personas per project. Hjarni's folder-level instructions let each persona behave the way that project needs, without duplicating notes.

Use the native MCP transport, no extra moving parts

Recent Open WebUI versions speak Streamable HTTP MCP directly from the admin panel. No proxy, no glue script: paste the Hjarni URL and the tools show up.

Mix self-hosted models with cloud assistants

The same Hjarni notes are read by Open WebUI today and Claude or ChatGPT tomorrow. Switching the model never requires migrating the knowledge.

Setup

Connect Open WebUI in about two minutes

  1. 1

    Sign up for Hjarni and have Open WebUI running with at least one model configured. Native MCP support requires a recent Open WebUI release — confirm in About if you are unsure.

  2. 2

    Sign in as an admin and open Admin Settings → External Tools. Click Add and pick the MCP (Streamable HTTP) option.

  3. 3

    Set the name to hjarni and the URL to https://hjarni.com/mcp. Open WebUI will surface an OAuth step on save when the server requires it.

  4. 4

    If your build does not list MCP as a server type, fall back to the mcpo bridge: pip install mcpo, then `mcpo --port 8000 --server-type streamable-http -- https://hjarni.com/mcp`, and add http://localhost:8000 as an OpenAPI tool server.

  5. 5

    Start a chat, ask the model to search Hjarni for one of your notes, and confirm the tool calls succeed.

Open WebUI → Admin Settings → External Tools → Add an MCP (Streamable HTTP) server

Type:  MCP (Streamable HTTP)
Name:  hjarni
URL:   https://hjarni.com/mcp
Auth:  OAuth (sign in with your Hjarni account when prompted)

The Hjarni MCP endpoint is https://hjarni.com/mcp. The connection uses OAuth, so you sign in with your Hjarni account when prompted. No API key files to manage.

Why Open WebUI plus Hjarni keeps your stack self-hosted on both sides

The reason to run Open WebUI is control: you pick the model, you pick where it runs. The matching choice on the knowledge side is to keep the notes outside any cloud chat product so the same context works whether you front it with Open WebUI today or a different UI tomorrow. Hjarni is that layer.

Open WebUI now speaks MCP natively in the External Tools panel, so the integration is a URL paste rather than a proxy. The mcpo bridge stays useful for stdio-only MCP servers and for older Open WebUI releases, but it is no longer the default for HTTP MCP endpoints like Hjarni.

Open WebUI is open source and your model can be too. Hjarni's notes are plain Markdown, exportable anytime. If you replace any one part of the stack, the rest keeps working.

Common questions

Questions before you connect Open WebUI

Does Open WebUI speak MCP natively?

Recent Open WebUI releases do. Admin Settings → External Tools lets you add a server with type MCP and transport Streamable HTTP, which is what Hjarni exposes. Older builds need the mcpo OpenAPI bridge instead.

When do I still need mcpo?

When the Open WebUI version you run does not list MCP as an External Tools server type, or when the MCP server you want to expose only speaks stdio. For Hjarni on a current Open WebUI build, mcpo is optional, not required.

How do I install mcpo if I need it?

Anywhere you can run Python. The simplest path is on the same machine that runs Open WebUI: pip install mcpo, then run it on a port Open WebUI can reach. In Docker setups, mcpo runs as a sidecar container on the same network.

Does this work with Ollama-only setups?

Yes. External Tools in Open WebUI are model-agnostic. Whether the assistant is Ollama, an OpenAI-compatible endpoint, or an Anthropic-compatible one, the same Hjarni MCP server is available.

Will Hjarni see my model or only my notes?

Hjarni only sees the MCP calls the model makes: searches, reads, and writes against your notes. Your conversation, model choice, and any other tool calls stay in Open WebUI.

Give Open WebUI a memory

The session starts where the last one ended. Write notes once. Open WebUI reads them across every conversation.

Get started free See pricing

More integrations

Hjarni for Claude Hjarni for ChatGPT Hjarni for Cursor Hjarni for Claude Code Hjarni for Windsurf Hjarni for Cline Hjarni for GitHub Copilot Hjarni for Zed Hjarni for Continue.dev Hjarni for Aider Hjarni for Raycast Hjarni for Alfred Hjarni for Obsidian

Write once. You both remember.

Free to start. No credit card required.

Give your AI a memory

Works with Claude and ChatGPT today. Gemini coming soon.