Using Scaleway Models in this Home Manager Configuration

This document explains how to use Scaleway models (Devstral) with the pi coding agent configured in this home-manager setup.

Prerequisites

Before using Scaleway models, you need to:

Have a Scaleway account with API access
Obtain your Scaleway Access Key and Secret Key
Store these credentials in the appropriate SOPS-encrypted files

Setting Up Scaleway Credentials

1. Update the Scaleway Secrets File

The Scaleway credentials are stored in secrets/scaleway.yaml and encrypted with SOPS. To edit:

SOPS_AGE_KEY_FILE=~/.config/sops/age/keys.txt sops secrets/scaleway.yaml

The file should contain:

access_key: "your-scaleway-access-key-here"
secret_key: "your-scaleway-secret-key-here"

2. Verify Model Configuration

Check that the Scaleway models are properly configured in pi-models.json:

{
  "providers": {
    "scaleway": {
      "baseUrl": "https://api.scaleway.ai/v1",
      "api": "openai-completions",
      "apiKey": "$SCW_SECRET_KEY",
      "compat": {
        "supportsDeveloperRole": false,
        "maxTokensField": "max_tokens",
        "requiresToolResultName": true,
        "supportsUsageInStreaming": false
      }
    }
  }
}

Note: The actual model definitions (including devstral-2-123b-instruct-2512 and qwen3-coder-30b-a3b-instruct) are now defined in pi-extensions/scaleway.ts rather than in pi-models.json.

Using Scaleway Models with pi

1. Accessing Models in pi

Once configured, the Scaleway models will be available in pi’s model selection interface:

“Devstral 2 123B (Scaleway)” - devstral-2-123b-instruct-2512 (Default)
“Qwen3 Coder 30B (Scaleway)” - qwen3-coder-30b-a3b-instruct
“Mistral Medium 3.5 128B (Scaleway)” - mistral-medium-3.5-128b (256k context window)

2. Model Selection in pi

When using pi, you can select these models through:

The /models command in pi to view available models
The model dropdown menu in the UI
Command-line specification when starting pi

3. Model Usage Restrictions

According to the agent configuration, Scaleway models:

Support text input only
Have large context windows (131,072 tokens for Devstral 2 123B and Qwen3 Coder 30B, 262,144 tokens for Mistral Medium 3.5 128B)
Support up to 16,384 max tokens (Scaleway API limit)
Do not support reasoning features
Are configured with zero cost (for tracking purposes)
Devstral 2 123B is now the default model
Mistral Medium 3.5 128B is available as an alternative model

Troubleshooting

1. Authentication Issues

If you encounter authentication problems:

Ensure your Scaleway credentials are correct in secrets/scaleway.yaml

Re-encrypt the file after making changes:

SOPS_AGE_KEY_FILE=~/.config/sops/age/keys.txt sops encrypt --overwrite secrets/scaleway.yaml

Reactivate the configuration:
```
home-manager switch --flake .#philipp
```

2. Model Availability

If the Scaleway models don’t appear in pi:

Check that the pi-models.json file is correctly formatted
Verify that the Ollama service is running (though Scaleway models don’t use Ollama)
Ensure the Scaleway API keys are properly set in the SOPS-encrypted file
Restart the pi agent or reload the configuration

Cost Considerations

The Scaleway models are configured with zero cost in the pi-models.json file. This is for tracking purposes only. Actual usage will incur charges according to Scaleway’s pricing for their API.

Testing Model Access

You can test if the Scaleway models are properly configured by:

Starting pi with the configuration
Using the /models command to list available models
Attempting a simple chat with one of the Scaleway models to ensure connectivity

Search Documentation