Chat

The Chat page provides a full-featured conversational interface for LLM interactions. Navigate to Chat in the sidebar or go to /chat.

On mobile:

Start a conversation

1. Select a model from the dropdown in the top bar
2. Type a message and press Enter
3. The model streams its response in real time

If no model is selected, you’ll be prompted to choose one. Your preferred model is remembered for new conversations.

Select a model

The model dropdown shows all available chat models with their status:

• Online — responds immediately
• Warming up — spinning up instances; the UI shows a countdown and retries automatically
• Standby — ready to activate
• Offline — unavailable

If a model is warming up, your request queues in the background and completes once the model is ready.

Attach images

Chat supports vision — attach images for models that support visual input:

1. Click the paperclip icon or drag and drop images onto the input
2. Thumbnails appear above the text field
3. Type your question about the images and send

Supports JPEG, PNG, GIF, and WebP up to 10 MB each, with a maximum of 5 images per message.

Tune model parameters

Click the settings icon (or press Cmd+Shift+S) to reveal the settings panel:

• System prompt — instructions prepended to every request in this conversation
• Temperature (0–2) — controls randomness; lower is more deterministic
• Max tokens — cap on response length; leave empty for auto
• Top P (0–1) — nucleus sampling threshold

Settings are saved per conversation. Click Reset all to return to defaults.

Use conversation controls

While the model is responding:

• Cancel — click the stop button or press Escape to abort

After a response completes:

• Copy — copy the response text
• Regenerate — re-send your last message to get a different response (available on the last assistant message)
• Retry — appears on error messages; re-sends the preceding user message
• Edit — click the pencil icon on any user message to rewrite it; all subsequent messages are removed and the conversation continues from your edit (press Enter to save, Escape to cancel)
• Read aloud — plays the response via text-to-speech if a TTS model is available

Export as code

Click the code icon in the top bar to open the View Code dialog. It generates ready-to-use snippets for your current conversation in three formats:

• curl — shell command with Bearer token auth
• Python — using the OpenAI SDK with base_url="https://api.casola.ai/openai/v1"
• TypeScript — using the OpenAI Node.js SDK

Snippets include your selected model, system prompt, and any non-default parameters. Copy and use them directly in your application.

API usage

You can use the chat completion endpoint directly from your application. The API is OpenAI-compatible, so existing SDKs work out of the box.

Basic request

curl https://api.casola.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3.5-4B",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is the capital of France?"}
    ],
    "temperature": 0.7,
    "max_tokens": 256
  }'

Response:

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1711234567,
  "model": "Qwen/Qwen3.5-4B",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "The capital of France is Paris."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 8,
    "total_tokens": 32
  }
}

Streaming

Set "stream": true to receive tokens as they’re generated via server-sent events:

curl https://api.casola.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3.5-4B",
    "messages": [{"role": "user", "content": "Write a haiku about clouds"}],
    "stream": true
  }'

Each event contains a delta with the next token:

data: {"choices":[{"delta":{"content":"Soft"},"index":0}]}
data: {"choices":[{"delta":{"content":" white"},"index":0}]}
...
data: [DONE]

Multi-turn conversation

Include the full conversation history in the messages array:

curl https://api.casola.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3.5-4B",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is 2+2?"},
      {"role": "assistant", "content": "2+2 equals 4."},
      {"role": "user", "content": "And what is that times 3?"}
    ]
  }'

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.casola.ai/openai/v1",
    api_key="YOUR_API_TOKEN",
)

# Streaming
stream = client.chat.completions.create(
    model="Qwen/Qwen3.5-4B",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

TypeScript (OpenAI SDK)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.casola.ai/openai/v1",
  apiKey: "YOUR_API_TOKEN",
});

const stream = await client.chat.completions.create({
  model: "Qwen/Qwen3.5-4B",
  messages: [{ role: "user", content: "Hello!" }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content || "");
}

Navigate conversations

The sidebar lists all your conversations, sorted by most recent. Each entry shows the conversation title and message count.

• New chat — start a fresh conversation (also Cmd+Shift+N)
• Rename — click to edit the title inline
• Delete — remove a conversation permanently

Conversations are persisted server-side and sync to your Library. On mobile, the sidebar collapses into a drawer accessible via the menu button.

Thinking/reasoning display

Some models include reasoning steps in their responses. These appear as collapsible “Show thinking” blocks — expandable to see the model’s chain of thought, separate from the final answer.