Mistral

The Mistral provider uses provider key mistral and the HPD-Agent.Providers.Mistral package. ModelName is a Mistral model id. This provider targets net10.0.

Set an API key:

export MISTRAL_API_KEY="..."

Use fluent setup first:

using HPD.Agent;
using HPD.Agent.Providers.Mistral;

var agent = await new AgentBuilder()
    .WithMistral(model: "mistral-large-latest")
    .BuildAsync();

var result = await agent.RunAsync("Write one sentence about Mistral setup.");
Console.WriteLine(result.Text);

The equivalent Clients.Chat shape is:

{
  "Clients": {
    "Chat": {
      "ProviderKey": "mistral",
      "ModelName": "mistral-large-latest"
    }
  }
}

Runtime Chat Options

Use ChatClientConfig for shared model-call behavior such as temperature, top-p, max output tokens, stop sequences, seed, response format, tools, tool mode, and generic reasoning options. Put agent-level defaults in AgentConfig.Clients.Chat; use AgentRunConfig.Clients.Chat for complete per-run replacements.

Mistral reasoning models use the generic reasoning setting:

var agent = await new AgentBuilder()
    .WithMistral(model: "mistral-large-latest")

    .BuildAsync();

Mistral Request Options

Use MistralChatRequestOptions for Mistral-specific per-request fields that are not generic chat options.

Option	Purpose
SafePrompt	Enables Mistral safety prompt injection before the conversation.
PredictionContent	Sends expected completion content for predictable edits and lower latency.
PromptCacheKey	Sets Mistral's prompt cache key for reusing prompt prefixes.
CompletionCount	Requests multiple completions for a single prompt.

var agent = await new AgentBuilder()
    .WithMistral(model: "mistral-large-latest")
    .WithMistralChatRequestOptions(mistral =>
    {
        mistral.SafePrompt = true;
        mistral.PromptCacheKey = "workspace-thread-1";
    })
    .BuildAsync();

Per-run configuration uses the same type:

var run = new AgentRunConfig
{
    Clients = new AgentClientsConfig
    {
        Chat = new ChatClientConfig
        {
            ProviderKey = "mistral",
            ModelName = "mistral-large-latest",
            ProviderOptions = new MistralChatRequestOptions
            {
                PredictionContent = "expected replacement text"
            }
        }
    }
};

Caveats

Validation checks model configuration. API key resolution happens during provider construction through MISTRAL_API_KEY, explicit apiKey, or configured secrets. Runtime chat options are applied through Clients.Chat and AgentRunConfig.Clients.Chat. Provider metadata reports streaming, function calling, vision input, and audio input support.