Chat & Testing

The Test page is the Designer's multi-purpose testing interface. It supports several model types and includes a built-in DevTools panel for debugging.

Model Types

The Test page supports switching between different model types via a selector:

Mode	What it tests
Inference	Standard chat with your configured LLM + optional RAG
Anomaly	Score inputs against trained anomaly detection models
Classifier	Classify text using trained classifier models
Document Scanning	OCR and document analysis
Encoder	Test embedding models and reranking
Speech	Voice interaction (see Speech)

Chat (Inference Mode)

The primary chat interface for testing your AI project:

• Streaming responses — see tokens appear in real-time
• RAG toggle — enable/disable document retrieval
• Source references — see which documents were retrieved and their relevance scores
• Markdown rendering — responses render with full markdown support (tables, code blocks, etc.)
• Voice input — click the mic button to speak instead of type
• Show prompts — toggle to see the full prompt sent to the model
• Show thinking — toggle to see chain-of-thought reasoning (if supported)

Session Management

• Chat history persists between visits
• Clear chat to start fresh
• Sessions are per-project

DevTools Panel

The DevTools drawer shows the raw API requests and responses for every interaction — invaluable for debugging and for building integrations.

Request Tab

See the exact HTTP request sent to the server:

• Method, URL, headers
• Request body (JSON)
• Timestamp and duration

Response Tab

See the full server response:

• Status code
• Response body
• Token usage (if available)

Code Snippets

The DevTools automatically generate code snippets for replicating the request in:

• cURL
• Python (requests)
• JavaScript (fetch)
• And more

Copy and paste these into your application to replicate exactly what the Designer did.

WebSocket Monitoring

For streaming responses, DevTools also tracks WebSocket connections with message-by-message logging.

Usage

• The DevTools panel appears as a collapsed bar at the bottom of the Test page
• Click to expand into a full panel
• Click outside or press Escape to collapse
• Clear history to reset captured requests

Document Scanning Mode

Test OCR and document analysis:

• Select a scanning backend
• Choose language
• Upload or paste document images
• View extracted text and structure

Encoder Mode

Test embedding and reranking models:

• Embeddings — enter text and see the vector output, test with common embedding models
• Reranking — provide a query and documents, see reranked results with scores

Sample inputs are provided for both to get started quickly.

API Routes

Action	Method	Route
Chat completion	POST	/v1/projects/{ns}/{project}/chat/completions
Chat (streaming)	POST	/v1/projects/{ns}/{project}/chat/completions (SSE)
Extract document	POST	/v1/vision/documents/extract
Create embeddings	POST	/v1/nlp/embeddings
Rerank documents	POST	/v1/nlp/rerank

Route

/chat/test