Send a chat message

What it does

Opens the chat panel and sends a message to the active model. Messages stream in real-time and are persisted in the conversation store. You can pick an existing conversation or create a new one, type your prompt, hit Enter, and watch the response appear token by token.

Prerequisites

• A running stack (the api container is healthy).
• At least one model is loaded (check Catalog to see available models).
• You know your active model id (shown in the model dropdown in the chat panel header).

Steps

1. Open the chat panel

Click the chat bubble icon in the PageHeader top-right corner, or press ⌘K (macOS) / Ctrl+K (Windows/Linux). The right-side panel slides in with the message list and input bar.

2. Pick a conversation

The panel opens to the active conversation. To switch:

• Click the conversation dropdown (shows the current conversation name).
• Select an existing one from the list, or click New conversation to create one.

3. Type your message

Click in the input bar at the bottom of the panel and type your prompt. Use Shift+Enter for multi-line input.

4. Send the message

Press Enter to send. The message is added to the list as a USER card, and a ASSISTANT card streams in below it.

5. Watch the stream

The assistant's reply appears token by token. Latency is shown when the stream completes (e.g. "1.2 s").

(Direct API) Send via curl

curl -s -X POST \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $BIGWAVE_TOKEN" \
  -d '{
    "model": "qwen2.5:3b",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is 2 + 2?"}
    ],
    "stream": true
  }' \
  "http://localhost:8080/v1/chat/completions" | jq

Response streams as newline-delimited JSON with delta.content for each chunk.

Verify

• Send a message with a system prompt and user turn.
• The assistant's reply appears in the chat panel within 1-2 seconds.
• Refresh Observability — your new inference is the top row.
• Open the row; the Context window drawer shows the SYSTEM + USER cards.

Common issues

Symptom	Cause	Fix
"Model not found"	No models are loaded or active model is null	Check Catalog to load a model, or use GET /v1/models to list what's available
Chat panel doesn't open	Keyboard shortcut not working in your browser	Try clicking the chat icon in PageHeader instead of using ⌘K
Message sends but no response	Model is loading (first request) or inference failed	Wait 10-20s for Ollama to load the GGUF; check api container logs for 5xx errors
"401 Unauthorized" on curl	Bearer token is missing or incorrect	cat .env | grep BIGWAVE_TOKEN and paste the correct value

Related

• Inspect inferences (Observability page) — view the full context window of each chat.
• Capture user feedback (👍/👎) — rate responses after they appear.
• Tool calls inspector — inspect calls if the model supports tool calling.