Do I need a new SDK to call Elevence AI from my existing OpenAI code?

No. The /v1/chat/completions endpoint is OpenAI-compatible. Keep your existing Python `openai` or Node `@openai/sdk` client, change `base_url` to https://api.elevence.ai/v1 and pass your Elevence key. Streaming, function calling, vision and multi-turn behave identically.

Which models can I call from one API key?

All 60+ models in the catalog — GPT-5, Claude 4.7 Opus & Sonnet, Gemini 3 Pro & Flash, Llama 4, Mistral, Cohere, Perplexity Sonar — plus the image, video and audio endpoints. The model identifier in the request decides which provider serves it.

Are tool calls / function calls supported?

Yes. The OpenAI tool-call shape is forwarded to the underlying model. GPT-5 and Claude support function calling natively; for models that don't, Elevence AI surfaces a clear error rather than silently degrading.

How does streaming work?

Standard OpenAI-style SSE. Pass `stream: true` and read the `text-delta` / `message-id` events. The same code that streams from OpenAI streams from Elevence AI without changes.

DEVELOPERS

OpenAI-compatible. 60+ models behind one key.

If your code already talks to /v1/chat/completions, it already talks to Elevence AI. Swap the base URL and the key — that's it.

Applies to web, iOS and Android

What you get

•

Drop-in /v1/chat/completions — same request shape, same SSE stream, same finish reasons.

•

One key, every model — switch by changing the model field.

•

Native multimodal — send images, PDFs, audio in the same request shape.

•

Reasoning effort — set reasoning_effort on supporting models.

•

Tool calls returned in the OpenAI shape, with our extended tools surfaced as friendly names.

Beyond OpenAI shape

•

Image / video / audio generation under /api/v1/generation/.

•

Threads, branches, forks for stateful conversations.

•

Share links, attachments uploaded to R2 with stable URLs.

•

Per-key usage and cost analytics.

Get a key

Sign in, then go to Settings → API Keys and create one. The key is shown once on creation and stored bcrypt-hashed on our side, so save it somewhere safe.

Read the docs

See /docs for the request / response reference and full provider matrix.