Controlled trial

text.generate: one stable contract.

Generate a draft or synthesis from a supplied prompt and target language. It does not browse, publish or access private accounts.

Prompt: up to 4,000 characters. Output: limited to 800 tokens. User and platform daily trial quotas apply.

Try the gateway ↗

Input

{"prompt":"Your approved brief","language":"Japanese"}

Output

{"text":"...","language":"Japanese","usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}

Protocol field names and endpoint IDs remain in English across all interface languages.

Authentication and retry

Run requires a RelayLane key or authenticated workspace. Supply a unique Idempotency-Key, and keep it when retrying the same input. Different input with the same key is rejected.

Failure and availability

A failed or uncertain upstream attempt is not automatically retried. Read the saved run status. A trial limit can block execution before a model request is sent.

For agents → · OpenAPI →

Before you connect

Do I need to register with every API provider?

Not for the current managed tools and public-data trial. You authorize one RelayLane account. Private-account APIs will still need separate permission.

Are thousands of external APIs already connected?

No. Check the live registry for the current verified and executable counts. Candidates and different access protocols are not counted as working APIs.

Can every large-model product use this?

Use REST from an application, MCP from a compatible client, or function schemas from a tool-calling model. Some hosted clients do not accept custom MCP headers.

Will search engines or AI products find this automatically?

Public HTML, sitemaps and machine-readable documents make discovery possible. They do not guarantee indexing or recommendation.