microsoft/semantic-kernel · error · KernelException

Chat completions not found

Error message

Chat completions not found

What it means

Thrown by MistralClient.GetChatMessageContentsAsync when the non-streaming chat completion response is null, has a null Choices collection, or has zero choices. The connector cannot produce ChatMessageContent without at least one choice. It is a KernelException raised after the HTTP call succeeded but the payload was empty.

Source

Thrown at dotnet/src/Connectors/Connectors.MistralAI/Client/MistralClient.cs:73

        var endpoint = this.GetEndpoint(mistralExecutionSettings, path: "chat/completions");
        var autoInvoke = kernel is not null && mistralExecutionSettings.ToolCallBehavior?.MaximumAutoInvokeAttempts > 0 && s_inflightAutoInvokes.Value < MaxInflightAutoInvokes;

        for (int requestIndex = 1; ; requestIndex++)
        {
            var chatRequest = this.CreateChatCompletionRequest(modelId, stream: false, chatHistory, mistralExecutionSettings, kernel);

            ChatCompletionResponse? responseData = null;
            List<ChatMessageContent> responseContent;
            using (var activity = ModelDiagnostics.StartCompletionActivity(this._endpoint, this._modelId, ModelProvider, chatHistory, mistralExecutionSettings))
            {
                try
                {
                    using var httpRequestMessage = this.CreatePost(chatRequest, endpoint, this._apiKey, stream: false);
                    responseData = await this.SendRequestAsync<ChatCompletionResponse>(httpRequestMessage, cancellationToken).ConfigureAwait(false);
                    this.LogUsage(responseData?.Usage);
                    if (responseData is null || responseData.Choices is null || responseData.Choices.Count == 0)
                    {
                        throw new KernelException("Chat completions not found");
                    }
                }
                catch (Exception ex) when (activity is not null)
                {
                    activity.SetError(ex);

                    // Capture available metadata even if the operation failed.
                    if (responseData is not null)
                    {
                        if (responseData.Id is string id)
                        {
                            activity.SetResponseId(id);
                        }

                        if (responseData.Usage is MistralUsage usage)
                        {
                            if (usage.PromptTokens is int promptTokens)
                            {

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Inspect the raw response and any moderation/safety fields (enable logging) to see why no choice was returned.
2. Retry once for transient API faults.
3. Adjust the prompt or moderation settings if content was filtered.
4. Verify the model id is valid and your plan has access to it.

Example fix

// before
var result = await kernel.InvokePromptAsync(prompt, new(new MistralAIPromptExecutionSettings()));  // throws

// after (handle empty-completion gracefully)
try
{
    var result = await kernel.InvokePromptAsync(prompt, new(new MistralAIPromptExecutionSettings()));
}
catch (KernelException ex) when (ex.Message == "Chat completions not found")
{
    logger.LogWarning("Mistral returned no chat completions; prompt may be filtered.");
    // fallback or re-prompt
}

try
{
    var result = await kernel.InvokePromptAsync(prompt, new(new MistralAIPromptExecutionSettings()));
}
catch (KernelException ex) when (ex.Message == "Chat completions not found")
{
    logger.LogWarning("Mistral returned no chat completions; prompt may be filtered or model unavailable.");
    // fallback, re-prompt, or surface to user
}

When it happens

Trigger: The Mistral API returned a 2xx response with no choices (e.g. content filtered entirely, all candidates blocked, an empty completion, or an unexpected API-side issue).

Common situations: Safety/content filters rejecting the entire generation; misconfigured model id; API quota/availability returning a degenerate 200; transient API faults; prompt that triggers a moderation block.

Related errors

• Role must be one of: system, user, assistant or tool. {role}
• Run failed with status: `{response.Status}` for agent `{agen
• Unsupported Mistral model: {modelId}
• Prompt was blocked due to Gemini API safety reasons.
• Unexpected response from model

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/650654661d4286c5. Report an issue: GitHub.