One ck- key works with any OpenAI-compatible client. Point the base URL here, pick a model, done.
https://conduit-api.bifrostapi.net/v1Create an account in seconds — free starter credit included.
Grab a ck-… key on the API Keys page.
Use the snippet below in code or your editor.
Replace ck-YOUR_KEY with your key. Any OpenAI SDK works.
curl https://conduit-api.bifrostapi.net/v1/chat/completions \
-H "Authorization: Bearer ck-YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.5",
"messages": [{"role": "user", "content": "Hello!"}]
}'Most models use chat completions — image and audio use their own.
POST /v1/chat/completionsall text modelsPOST /v1/images/generationsgpt-image-2 · gemini-3-pro-image-preview → data[0].b64_jsonPOST /v1/chat/completionsgemini-3.1-flash-image → message.imagesPOST /v1/audio/speecheleven-flash · eleven-multilingual · qwen-tts · qwen3-tts-vc → input = the text; optional voicePOST /v1/audio/speecheleven-sfx → input = a description ("wooden door creak"); duration_seconds ≤ 30, loop for a seamless bedPOST /v1/audio/speecheleven-music → input = a description ("8-bit chiptune loop"); duration_seconds ≤ 600, force_instrumental for no vocalsPOST /v1/music/referencesmultipart: model=eleven-music-reference, file=@clip.mp3 → reference_id, then pass it to /v1/audio/speechPOST /v1/audio/transcriptionsdeepgram-nova-3 · speechmatics-*model-high / -low / -xhigh suffix, or the reasoning_effort param (gpt-5.x / o3 / claude). Gemini uses -low / -agent variants.Drop these into your favorite AI client.
~/.codex/config.tomlmodel = "gpt-5.5" model_provider = "bifrostapi" [model_providers.bifrostapi] name = "BifrostAPI" base_url = "https://conduit-api.bifrostapi.net/v1" wire_api = "responses" requires_openai_auth = false experimental_bearer_token = "ck-YOUR_KEY"
Tip — after signing in, Integrations generates a key and fills these in with one click.