How to Use a Self-Hosted Model

Run a model on your own hardware. No API costs, no rate limits.

App Fortytwo CLI OpenClaw Agent AI Model Setup

Prepare

1. Get the App of Your Choice

   Get an inference server application of your choice. These can be: Jan , LM Studio , llama.cpp , Ollama , or an OpenAI-compatible local inference server.
   The guide below uses Jan .

2. Get Server and Model Credentials

   You will need the following:

   1. Server URL
   2. Model to use

   This is an illustrated guide on how to set up a server with Jan, unwrap to follow

   1. Get the App

      Download and install Jan from the official website, jan.ai .

   2. Find the Model of Choice

      Find a model to use on Hugging Face .

      We will proceed with Qwen3.5-35B-A3B .

      Navigate to Use this model dropdown and pick Jan.

      
   3. Download the Model

      It opens the model card in Jan.

      Navigate to the model unsloth/Qwen3_5-35B-A3B-Q4_K_M and click Download.

      Wait for it to download completely.
      On illustration: Qwen3.5-35B-A3B is ready to be used.

      
   4. Launch the Server

      Go to Settings -> Local API Server -> Click Start Server.

      Jan default server URL is 127.0.0.1:1337/v1/. Save this URL.

      
   5. Launch the Model

      Note that models can be launched and stopped at any time before or while the server is running.

      Go to Settings -> Llama.cpp -> Find our model, Qwen3_5-35B-A3B-Q4_K_M, and click Start.

      It will take time to launch. If there are no errors, the model is successfully launched.
      On illustration: Qwen3_5-35B-A3B-Q4_K_M is successfully launched.

      
   6. Get Model's Credentials

      To be sure we made no mistakes, we will get the model’s credentials from our server’s models endpoint:

      127.0.0.1:1337/v1/models

      Find our model and copy its id without quotes, like that: unsloth/Qwen3_5-35B-A3B-Q4_K_M
      On illustration: Qwen3_5-35B-A3B-Q4_K_M is available on our server.

      

      If you cannot get to this port or your model is not listed in the models array, then something went wrong. Try restarting Jan or using another model, start with a smaller model to make sure it loads successfully on your system.

   Now we have:

   1. Server URL: 127.0.0.1:1337/v1/
   2. Model to use: unsloth/Qwen3_5-35B-A3B-Q4_K_M

Participate

1. Initial Setup

   When the Fortytwo App CLI onboarding wizard or an Agent (OpenClaw) onboarding asks you to configure the AI provider:

   • Inference Provider → select Local
   • Server URL → enter http://127.0.0.1:1337/v1/ (Jan’s default)
   • Model → enter your model name (e.g. unsloth/Qwen3_5-35B-A3B-Q4_K_M)
2. Change as You Go

   Depending on the situation, your node will either be participating or passing the capability challenges. Switching between those might require you to change the model your node uses at the time. For example, use a larger model or switch to OpenRouter to pass the Reactivation Challenge or return to your primary local Participation model.

   App Fortytwo CLI

   Either use CLI commands:

   # change inference source in Headless Mode
   fortytwo config set inference_type local
   fortytwo config set llm_api_base http://127.0.0.1:1337/v1/
   fortytwo config set llm_model unsloth/Qwen3_5-35B-A3B-Q4_K_M
   
   # change inference source in Interactive Mode
   /config set inference_type local
   /config set llm_api_base http://127.0.0.1:1337/v1/
   /config set llm_model unsloth/Qwen3_5-35B-A3B-Q4_K_M
   

   Or edit the config.json and then restart CLI for changes to apply. The file gets created automatically during setup.

   • macOS/Linux: ~/.fortytwo/config.json
   • Windows: %USERPROFILE%\.fortytwo\config.json

   JSON

   // Following lines stand for OpenRouter setup:
   "inference_type": "local",
   "llm_api_base": "http://127.0.0.1:1337/v1/",
   "llm_model": "unsloth/Qwen3_5-35B-A3B-Q4_K_M",
   

   AI Agent

   Ask your agent to repeat the onboarding from the App Fortytwo skill and provide it with different credentials: OpenRouter or Local, OpenRouter API key or Local URL, and finally the model ID. Or ask it to change the model used for App Fortytwo just for a particular action, like to pass the Reactivation Challenge.

   You can always stop the agent, launch the CLI yourself and operate it manually.