Mistral AI Provider Plugin

The Mistral plugin connects Ever Works to Mistral AI's API, providing access to high-performance language models optimized for efficiency, speed, and multilingual tasks. It extends BaseAiProvider and uses the shared AiOperations layer that wraps LangChain under the hood.

Source: packages/plugins/mistral/src/mistral.plugin.ts

Overview

Property	Value
Plugin ID	mistral
Package	@ever-works/mistral-plugin
Category	ai-provider
Capabilities	ai-provider
Version	1.0.0
Configuration Mode	user-required
Provider Type	mistral
Auto-enable	No
Built-in	Yes
Visibility	public

The plugin extends BaseAiProvider from @ever-works/plugin/abstract and creates an AiOperations instance configured with providerType: 'mistral' to handle all AI operations through the unified LangChain-based abstraction.

Architecture

Mistral's API is OpenAI-compatible, so the plugin uses the same LangChain OpenAI provider with a custom baseURL pointing to https://api.mistral.ai/v1.

Configuration

Settings Schema

Setting	Type	Required	Default	Scope	Widget	Description
apiKey	string	Yes	--	user	--	Mistral API key. Secret. Env: PLUGIN_MISTRAL_API_KEY
defaultModel	string	Yes	mistral-small-latest	global	model-select	Default model for all AI tasks. Env: PLUGIN_MISTRAL_DEFAULT_MODEL
simpleModel	string	No	mistral-small-latest	global	model-select	Model for tags, descriptions, classifications.
mediumModel	string	No	mistral-medium-latest	global	model-select	Model for listings, summaries, reformatting.
complexModel	string	No	mistral-large-latest	global	model-select	Model for full page generation, multi-step analysis.
temperature	number	No	0.7	--	--	Sampling temperature (0--2). Hidden.
maxTokens	number	No	4096	--	--	Max tokens per response. Hidden.
baseUrl	string	No	https://api.mistral.ai/v1	--	--	API endpoint. Hidden. Env: PLUGIN_MISTRAL_BASE_URL

Model Tiers

The plugin supports a tiered model system that maps task complexity to appropriate models:

Tier	Default Model	Use Cases
Simple	mistral-small-latest	Tags, short descriptions, quick classifications
Standard	mistral-medium-latest	Listings, summaries, content reformatting
Complex	mistral-large-latest	Full page generation, multi-step analysis

Environment Variables

Variable	Description
PLUGIN_MISTRAL_API_KEY	Mistral API key (overrides UI setting)
PLUGIN_MISTRAL_DEFAULT_MODEL	Default model ID
PLUGIN_MISTRAL_SIMPLE_MODEL	Simple tier model ID
PLUGIN_MISTRAL_MEDIUM_MODEL	Medium tier model ID
PLUGIN_MISTRAL_COMPLEX_MODEL	Complex tier model ID
PLUGIN_MISTRAL_BASE_URL	Custom API endpoint

Capabilities

The getCapabilities() method reports the following:

Capability	Supported
Structured Output	Yes
Streaming	Yes
Tool Calling	Yes
Vision	Yes
Max Context Length	128,000 tokens

Supported Operations

// Chat completion
const response = await mistralPlugin.createChatCompletion({
	messages: [{ role: 'user', content: 'Describe this restaurant' }],
	settings: { apiKey: 'your-key' }
});

// Streaming chat completion
for await (const chunk of mistralPlugin.createStreamingChatCompletion({
	messages: [{ role: 'user', content: 'Write a description' }],
	settings: { apiKey: 'your-key' }
})) {
	process.stdout.write(chunk.content ?? '');
}

// Embeddings
const embedding = await mistralPlugin.createEmbedding({
	input: 'semantic search text'
});

// Structured JSON output
const result = await mistralPlugin.askJson('Generate categories', {
	settings: { apiKey: 'your-key' }
});

// List available models
const models = await mistralPlugin.listModels({
	apiKey: 'your-key'
});

Configuration Mode

The Mistral plugin uses user-required configuration mode, meaning each user must provide their own API key. There is no admin-level shared key option.

Settings Resolution

When an AI operation is called, settings are resolved using the resolveConfig method:

protected override resolveConfig(settings?: PluginSettings): Partial<AiOperationsConfig> {
    const s = settings ?? {};
    const config: Partial<AiOperationsConfig> = {};

    if (s.apiKey && typeof s.apiKey === 'string') {
        config.apiKey = s.apiKey;
    }
    if (s.baseUrl && typeof s.baseUrl === 'string') {
        config.baseURL = s.baseUrl;
    }
    if (typeof s.temperature === 'number') {
        config.temperature = s.temperature;
    }
    if (typeof s.maxTokens === 'number') {
        config.maxTokens = s.maxTokens;
    }
    return config;
}

The resolved config is passed to AiOperations as overrides, taking precedence over the defaults set during onLoad.

Lifecycle

Health Check

The plugin reports a simple health check that always returns healthy when the plugin class is instantiated:

async healthCheck(): Promise<PluginHealthCheck> {
    return {
        status: 'healthy',
        message: 'Mistral plugin is ready',
        checkedAt: Date.now()
    };
}

To verify actual API connectivity, use isAvailable() with a valid API key, which calls testConnection() against the Mistral API.

Dependencies

Package	Version	Purpose
@ever-works/plugin	workspace	Plugin contracts and base classes
@langchain/openai	^0.6.17	LangChain OpenAI-compatible provider
@langchain/core	^0.3.80	LangChain core abstractions

Getting Started

1. Create an account at console.mistral.ai
2. Generate an API key from the Mistral console
3. Enable the Mistral plugin in the Ever Works plugin settings
4. Enter the API key in the Mistral API Key field
5. Select preferred models for each task complexity level