# The AI-native knowledge base for Jan

Jan runs models privately on your own machine. Hjarni gives that local model a memory: durable notes it can read at query time without sending your data to a model vendor.

25 notes free, forever. No credit card required.

## Connect Jan in about two minutes

1. Sign up for Hjarni and make sure Node.js is installed, since the bridge runs through npx.
2. In Jan, open Settings, then MCP Servers, and add a new server.
3. Name it hjarni, set Command to npx, and set Arguments to -y mcp-remote https://hjarni.com/mcp.
4. On first run the mcp-remote bridge opens a browser to authorize your Hjarni account. Enable tool calling and pick a model that supports tool calls.
5. Start a chat and ask the model to list your Hjarni notes to confirm the connection.

```
npx -y mcp-remote https://hjarni.com/mcp
```

Jan adds MCP servers as local commands (Settings, then MCP Servers, then +), with Name, Command, and Arguments fields, so a hosted server like Hjarni connects through the mcp-remote bridge. Set the command to `npx` and the arguments to `-y mcp-remote https://hjarni.com/mcp`. The bridge proxies Hjarni over stdio and opens a browser for the Hjarni sign-in on first run. Requires Node.js, and a model with tool calling enabled.

## Workflows that actually use your context

1. Memory for a model with none

   A local model in Jan has no built-in memory and no web context. Connect Hjarni and it can read your accumulated notes at query time, while the model stays on your machine.

2. Private by default, on both ends

   Jan is local-first. Hjarni stores your notes in the EU and only sends what a connected client requests. You keep the model local and choose what context it pulls.

3. Carry context between sessions

   Jan conversations do not remember each other. Write the important things to Hjarni once and the next session reads them back instead of starting cold.

4. Capture what you learn

   Hjarni's MCP tools include create and update, so you can ask Jan to save a result or decision as a note for later.

## Why a local-first model still wants an external knowledge base

Running a model locally in Jan solves privacy and cost. It does not solve memory. The model is the same blank slate every session, with no idea what you decided, what your conventions are, or what you already told it.

Hjarni is where that durable context lives, as Markdown notes you own and can export anytime. The model stays on your machine; only the specific notes it asks for travel over the connection when you choose to connect.

Jan adds MCP servers as local commands, so a hosted server like Hjarni connects through the mcp-remote bridge, which proxies it over stdio and handles the sign-in. The result is a private setup that still remembers across sessions.

Common questions

## Questions before you connect Jan

**Does Jan work with a remote MCP server like Hjarni?**

Yes, through a bridge. Jan adds MCP servers as local commands (Settings, then MCP Servers), so a hosted server like Hjarni connects via the mcp-remote bridge: command npx, arguments -y mcp-remote https://hjarni.com/mcp. Tool calling must be enabled and the model you use must support tool calls.

**How do I authenticate Jan to Hjarni?**

The mcp-remote bridge opens a browser to sign in to your Hjarni account on first run, so you do not paste a token into Jan. It needs Node.js, since the bridge runs through npx.

**Does my data leave my machine?**

Jan runs the model locally. Only the specific notes the model requests travel to it over the MCP connection, and only when you connect Hjarni. Notes are stored in the EU.

**Why connect a notes app to a local model?**

A local model has no built-in memory and no web context. Hjarni gives it durable notes to read at query time while the model itself stays on your machine.

**Is the connection free?**

Yes. Hjarni's free tier includes MCP access with no token limits, and Jan is free and open-source.