Playground

Test prompts, inspect responses, and see how Aivana handled each answer, all in your browser.

The Aivana Playground is a visual interface for testing prompts in real time. You can adjust parameters, attach images and documents, see how each answer came together, inspect the exact request and response, and copy ready-to-use code snippets, all without writing any code.

• Navigate to the Playground from the sidebar or top navigation
• Type your prompt in the message input
• Click "Send" or press Enter to get a response
• View the AI response in the chat panel

Enter any prompt and see the MMI response instantly. While an answer is on its way, the indicator shows how long it has been working and, once the answer streams in, about how many tokens it has written. The Intelligence Trace panel shows how the answer came together: how many independent perspectives contributed, and whether live sources were used. If you stop an answer partway, you're charged for your prompt and the part you received. What each request was charged is in Usage & Insights.

Open the Configure tab to set the controls below. Changes take effect on your next request. Parameters explains each option in the API.

• System Instructions: your own instructions, such as a persona, tone or format.
• Temperature: how varied the answer is.
• Max Tokens: a ceiling on answer length.
• Effort: a ceiling on how much intelligence Aivana may apply to the request.
• Response Format: Text, or Structured. Structured then offers Prompt-defined JSON (describe the fields you want in your prompt; best effort) or JSON Schema (Strict) (you define the exact structure and Aivana validates the response against it).
• Stop Sequences: strings where the answer should end.
• Web Search Allowed: lets Aivana use live web information when your question needs it. A request that searches uses more tokens.

Attach up to five supported images or documents to analyze, compare, or reference in a prompt. Aivana supports PNG, JPEG, WebP, GIF, PDF, DOCX, and CSV files. Use the paperclip button, paste files into the message box, or drag them onto the chat. See Files & Documents for supported limits and API usage.

Open the JSON tab to see the exact request the Playground sent and the response the API returned: what your own integration would send and receive. A streamed answer is shown as its server-sent events, with each run of text deltas merged into one entry. File data is shortened there, and auth headers are never shown.

To get the answer as JSON your application can parse, set Response Format to Structured in the Configure tab, then choose Prompt-defined JSON and name the fields you want in your prompt. The answer is shown as a formatted, copyable JSON block, and the API also returns it parsed in the structured field. Prompt-defined JSON is best effort. See Parameters.

To get an answer that matches a structure you define, choose Structured, then JSON Schema (Strict). A schema editor appears with a small starter schema; edit it, or choose Reset to example to start again. Aivana validates the response against your schema before returning it.

• The editor checks your schema as you type and shows what to fix. A schema Aivana can't accept stops the request before it is sent, and nothing is charged. Aivana checks it again when you send, and that check is the one that counts.
• A strict request is answered whole, so the response streaming switch is turned off while JSON Schema (Strict) is on, and comes back when you leave it.
• Stop Sequences can't be used with JSON Schema (Strict); clear them first.
• A successful response shows the validated object, formatted and copyable, with Schema validated beneath it. That mark means Aivana served your request against your schema. It never appears for Prompt-defined JSON, which is best effort.
• If Aivana can't produce a response that matches, you see an error instead of an answer, and you aren't charged.

The Code tab then gives you the equivalent request with your schema, in cURL, Python or Node.js, and the JSON tab shows the exact request and response. See Structured Outputs for a worked example, and Streaming for why a strict request isn't streamed.

Open the Code tab to get a ready-to-use snippet in cURL, Python or Node.js with your current settings. Copy it directly into your project to send the same request from your code.