Skip to content

Using PlanExe with Mistral

This is for advanced users that already have PlanExe working.

If you are getting started with PlanExe, I (the developer of PlanExe) recommend following the openrouter.md instructions. Stick with the default settings.

If you want to use Mistral, then OpenRouter has several mistral models.

Docker setup for Mistral

Mistral support is not baked into the Docker image by default. You must add the Mistral LlamaIndex extension to the worker, rebuild the image, and supply your API key.

1) Install Docker (with Docker Compose), then clone the repo and enter it: 

git clone https://github.com/PlanExeOrg/PlanExe.git
cd PlanExe
2) Enable the Mistral client inside the worker image by editing worker_plan/pyproject.toml. Under [project].dependencies, add or uncomment these lines: 
"llama-index-llms-mistralai==0.4.0",
"mistralai==1.5.2",
Without this step, the Docker image will not have the MistralAI class. 3) Copy .env.docker-example to .env and add your key: 
MISTRAL_API_KEY='INSERT-YOUR-SECRET-KEY-HERE'
4) Add (or keep) a Mistral entry in llm_config.json (example below). 5) Rebuild the images so the new dependencies are baked in: 
docker compose build --no-cache worker_plan frontend_single_user
6) Start PlanExe: 
docker compose up worker_plan frontend_single_user
7) Open http://localhost:7860, go to Settings, and pick your Mistral model (e.g., mistral-paid-large). If you later tweak only llm_config.json, just restart the containers (docker compose restart worker_plan frontend_single_user); rebuilds are only needed when dependencies change.

Why use Mistral?

Mistral can have run your own fine tuned model in the cloud. If you have sensitive business data that you don't want to share with the world, then this is one way to do it.

Create an account on the mistral.ai website and buy 10 EUR of credits.

List of available models.

Using the free models, and the API is rate limited to 1 request per second. PlanExe cannot deal with rate limiting and PlanExe does 70-100 requests, so it's likely going to yield errors.

Create API key

  1. Visit api-keys.
  2. Click Create new key and name the new key PlanExe.
  3. In the .env file in the root dir of the PlanExe repo, create a row named MISTRAL_API_KEY. Copy/paste the newly created api key into that row.

The .env file should look something like the following, with your own key inserted. 

MISTRAL_API_KEY='AWkg3SxFTLWaPJClbASfv9h3VPItroof'

Edit the llm_config.json

The JSON should look something like this:

{
    "mistral-paid-large": {
        "comment": "This is paid. Possible free to use for a limited time. Check the pricing before use.",
        "class": "MistralAI",
        "arguments": {
            "model": "mistral-large-latest",
            "api_key": "${MISTRAL_API_KEY}",
            "temperature": 1.0,
            "timeout": 60.0,
            "max_tokens": 8192,
            "max_retries": 5
        }
    }
}

Use the Mistral model

  1. Restart PlanExe
  2. Go to the Settings tab
  3. Select the mistral-paid-large model.