Chat and streaming

A chat completion

temp := 0.7
maxTokens := 400
reply, err := client.Chat.Completions.New(ctx, freetheai.ChatCompletionParams{
    Model:       "fta/zai/glm-5.3",
    Messages:    []freetheai.Message{{Role: "user", Content: "Name three uses for a paperclip."}},
    Temperature: &temp,
    MaxTokens:   &maxTokens,
})

Optional numbers are pointers, so an unset field is left out of the request. Any other Chat Completions field can go in Extra:

params.Extra = map[string]any{"frequency_penalty": 0.2}

Streaming

NewStreaming returns once the first piece of the reply has arrived. Read it with Next, then check Err:

stream, err := client.Chat.Completions.NewStreaming(ctx, freetheai.ChatCompletionParams{
    Model:    "fta/kimi/k3",
    Messages: []freetheai.Message{{Role: "user", Content: "Tell me a short story."}},
})
if err != nil {
    log.Fatal(err)
}
defer stream.Close()

for stream.Next() {
    for _, choice := range stream.Current().Choices {
        fmt.Print(choice.Delta.Content)
    }
}
if err := stream.Err(); err != nil {
    var stopped *freetheai.StreamInterruptedError
    if errors.As(err, &stopped) {
        fmt.Println("\nstopped early, text so far:", stopped.Partial)
    }
}

If you only want the text, StreamText calls your function for each piece:

err := client.Chat.Completions.StreamText(ctx, params, func(text string) error {
    _, err := io.WriteString(os.Stdout, text)
    return err
})

The last chunk carries Usage.

Reasoning

Thinking models return their reasoning in ReasoningContent. Turn thinking off with ReasoningEffort: "none". Text inside a leading <think> block moves to ReasoningContent, so Content holds only the answer.

Tool calls

tools := []freetheai.Tool{{
    Type: "function",
    Function: freetheai.ToolFunction{
        Name:        "get_weather",
        Description: "Current weather for a city",
        Parameters: map[string]any{
            "type":       "object",
            "properties": map[string]any{"city": map[string]any{"type": "string"}},
            "required":   []string{"city"},
        },
    },
}}
messages := []freetheai.Message{{Role: "user", Content: "Is it raining in Paris?"}}
reply, err := client.Chat.Completions.New(ctx, freetheai.ChatCompletionParams{
    Model: "fta/zai/glm-5.3", Messages: messages, Tools: tools,
})
// Run each reply.Choices[0].Message.ToolCalls entry, then append a role "tool" message
// with the same ToolCallID and send the conversation again.

Images in messages

Models that read images take content parts:

messages := []freetheai.Message{{
    Role: "user",
    Content: []freetheai.ContentPart{
        {Type: "text", Text: "What is in this picture?"},
        {Type: "image_url", ImageURL: &freetheai.ImageURL{URL: "https://example.com/cat.jpg"}},
    },
}}

A text-only model returns *ImagesNotSupportedError; with fallback models set, the next model is tried.