Setup¶

TLDR; The Endpoints

While we provide instructions below for several tools (npm, uv, opencode etc.). If you are in a hurry, here are the information about the inference and MCP endpoints for your reference.

Endpoint for the model: http://mimer-aif-hackathon.icedc.se/v1
Endpoint for Riksbanken MCP: http://mimer-aif-hackathon.icedc.se/swemo/sse
Endpoint for SCB/Statistics Sweden MCP: http://mimer-aif-hackathon.icedc.se/scb/mcp

We have also added a skills repo for the SCB MCP available here: https://github.com/ashwinvis/scb-opendata-skills

In this hackathon, we execute MistralAI’s Devstral 2 Small (256K, 24B) in H100 GPUs, served through multiple instances of vLLM. The model was chosen as it fits consumer hardware and therefore can simulating a home development for those who never did it.

Recommended Tools¶

The following tools are often useful for either installing MCP servers or building the final application for your hackathon.

Node.js via npm and npx commands: Follow instructions at https://nodejs.org/en/download
Python via uv and uvx commands: Follow instructions at https://docs.astral.sh/uv/getting-started/installation/
(Optional, but convenient to have) Git for cloning skill repositories instead of downloading. We use git clone command in instructions below. Follow instructions at https://git-scm.com/ to get git.

Note on virtual environments

We have prepared a guide on Python and use of virtual environments if you are unfamiliar with it.

Standard installation with open-weights LLM from Mimer¶

For this Hackathon, we support the usage of OpenCode as it is a free and opensource tool that do not require any third-party account. If you want to use other tools (Claude Code, Mistral Vibe, etc), you can directly refer to their documentation and configuration of custom endpoints.

Important

Please notice that while there are instructions for Windows, we strongly support using Windows Subsystem for Linux (see WSL in the next tab) as the development environment is more robust and secure (i.e., it is easier to administrate Python environments).

Download the OpenCode at its official download webpage.
Open OpenCode once, and close it.
On Windows Explorer, open %USERPROFILE%\.config\opencode
Create a file named opencode.jsonc, and insert the JSON configuration given below, replacing the entire content, if the file was not empty:
Now reopen OpenCode.

In your terminal, download OpenCode by running[1]

curl -fsSL https://opencode.ai/install | bash

Run opencode, and then type exit and hit Enter.
Execute the following to prepare the configuration file

mkdir -p ~/.config/opencode/
cd ~/.config/opencode/
touch opencode.jsonc

Then use your favourite text editor to replace the contents of opencode.jsonc, with the JSON configuration given below.
Run opencode and start using it!

(Optional): you can also run opencode web --port 4096 --hostname 0.0.0.0 and start a web instance of it in your browser, which will then be available on http://localhost:4096.

OpenCode JSON configuration

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "mimer": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Mimer AI",
      "options": {
        "baseURL": "http://mimer-aif-hackathon.icedc.se/v1"
      },
      "models": {
        "mistralai/Devstral-Small-2-24B-Instruct-2512": {
          "name": "Devstral 2 Small (256k)",
          "limit": {
            "context": 262144,
            "output": 262144
          }
        },
      }
    }
  },
  "model": "mistralai/Devstral-Small-2-24B-Instruct-2512",
  "mcp": {
    "scb_opendata_mcp": {
      "type": "remote",
      "url": "http://mimer-aif-hackathon.icedc.se/scb/mcp",
      "enabled": true
    },
    "swemo-mcp": {
      "type": "remote",
      "url": "http://mimer-aif-hackathon.icedc.se/swemo/sse",
      "enabled": true
    }
  }
}

Use Ctrl+P to access menus and switch the models and/or the MCPs. You can also use slash commands like /models and /mcps. You should now be able to select the model (Devstral 2 Small (256k)).

Skills¶

Agent skills can be added by running the commands:

Launch Git bash from the start menu. It should look like this. Then run the following commands.

$ mkdir -p $USERPROFILE/.config/opencode/skills

$ git clone https://github.com/ashwinvis/scb-opendata-skills
$ cp -r scb-opendata-skills/skills/*  $USERPROFILE/.config/opencode/skills/

$ mkdir -p ~/.config/opencode/skills

$ git clone https://github.com/ashwinvis/scb-opendata-skills
$ cp -r scb-opendata-skills/skills/*  ~/.config/opencode/skills/

Happy coding 🥳!