Mistral AI

Create Account and Obtain API Key

Register an account on La Plateforme of Mistral and create an API key.

Register at console.mistral.ai, go to API Keys, click Create new key and save the key into an environment variable.

Check the initial free tier limit on the dashboard before adding more to avoid interruptions during testing.

Install SDK and Test Basic API Call

Install the official client library and attempt to call a simple chat completion request.

Run `pip install mistralai`, then call `client.chat.complete(model="mistral-large-latest", messages=[...])`.

Use the 'mistral-small' model for simple tasks to save costs before needing 'mistral-large'.

Configure Function Calling for Business Tasks

Define internal functions (e.g., order lookup) for the model to call when needed.

Pass the `tools` parameter describing the function schema (name, parameters, description) in the request, handle the response when the model returns `tool_calls`.

Write the function description clearly in English in the schema, helping the model call the correct function even if the user asks in Vietnamese.

Enable JSON Output Mode for Structured Data

Force the model to return the correct JSON format when the application needs to parse data automatically.

Pass `response_format={"type": "json_object"}` in the request and clearly describe the desired JSON schema in the system prompt.

Always validate the returned JSON with code before using it, in case the model misses a field.

Implement Error Handling and Cost Limits

Add retry mechanisms and token limits to control costs and stability.

Wrap the request in a try/catch to handle rate limit errors, set `max_tokens` appropriately for each type of task.

Monitor weekly usage on console.mistral.ai to detect early if costs increase unusually.