Build a SchneeAI request. Copy runnable code.

The builder below writes the request for you. Pick a model (or auto), fill in the prompts, tune the sampling parameters, and switch tabs to see the same call in four languages. Copy drops a snippet into your clipboard — paste it into a terminal or your codebase and run.

Everything here is client-side. The form state is saved to your browser’s localStorage; we do not transmit it.

What each field does

• Model — the model alias SchneeAI routes to. auto lets the Gateway pick based on your routing policy. Pick Custom model to type a fine-tune or self-hosted alias verbatim.
• System prompt (optional) — prepended as a system message. Use it for persona, instructions, and output format. Omit entirely if your prompt is self-contained.
• User message — the primary instruction or question. Required; the placeholder <your instruction> is used if you leave it blank, so the snippet still parses.
• Temperature — sampling temperature, 0 to 2. Lower is more deterministic; above ~1.2 gets creative (and often worse).
• Max tokens — hard cap on completion length. Pair with temperature: 0 and a low cap for classification tasks.
• Stream — toggles stream: true. curl outputs raw SSE; the Python / TypeScript / Go snippets include the non-streaming parse path as the simpler default.

Patterns the snippets follow

• Auth via env var — every snippet reads $SCHNEEAI_API_KEY / process.env.SCHNEEAI_API_KEY / os.Getenv("SCHNEEAI_API_KEY"). Never inline tokens in source.
• OpenAI-compatible shape — request body matches OpenAI’s /v1/chat/completions. Anything you’d send to OpenAI works against SchneeAI unchanged.
• Explicit timeouts — 60s on Python, default http.DefaultClient in Go (override in production).
• Error handling left to you — the snippets raise/throw on non-2xx, but production code should retry on 429/500/502/503/504 with exponential backoff and jitter. See Error Handling.

When to reach for auto

"model": "auto" is the default for a reason. The Gateway’s routing policy resolves it per request based on your tenant configuration — feature tag, budget, latency target, PII policy. For exploration and most production traffic, auto is the right call. Pin a specific model when:

• You need deterministic capability (e.g., tool calling with Claude 3.5 Sonnet).
• You’re A/B testing models and need stable assignment.
• You’re hitting a budget cap and want a known-cheap option.

Next steps

• Tutorial — full walkthrough: streaming, named prompts, retries, cost tracking.
• Model Directory — every model the Gateway routes to, with effective pricing.
• Cost Calculator — estimate per-request and monthly spend before you ship.
• LLM Pricing Benchmark — compare per-token prices across providers.
• Prompt Templates — production-ready prompts for summarization, RAG, code review, extraction, and PII masking.