One OpenAI-compatible API for chat, image, video, speech, embeddings and more. Yelno routes each request under your organization's policies — hard constraints first, then price — and every response returns its route, usage and charge.
Chat, image, video, speech, transcription, embeddings, rerank and decision models through one OpenAI-compatible API — counted here from the live catalog.
Text out — chat, tools, structured output, long context.
POST /v1/chat/completionsPrompt in, images out, with reference images where the model takes them.
POST /v1/images/generationsAsynchronous jobs: submit, poll, stream the finished file.
POST /v1/videosText in, audio bytes out, in the format you ask for.
POST /v1/audio/speechAudio in, transcript out — segments and words where returned.
POST /v1/audio/transcriptionsText in, vectors out, for search and retrieval indexes.
POST /v1/embeddingsA query and documents in, relevance-ordered results out.
POST /v1/rerankContent and named questions in, typed answers out.
POST /v1/decisionsLive speech in, speech out, over one WebSocket session.
WS /v1/realtimeOpenAI-compatible with a base-URL change. Pick a model from the live catalog and read the fields every real response carries: the endpoint that served it, the attempts, the usage and the charge.
from openai import OpenAI
client = OpenAI(
base_url="https://yelno.ai/v1",
api_key=os.environ["YELNO_API_KEY"],
)
res = client.chat.completions.create(
model="MODEL",
messages=[{"role": "user", "content": prompt}],
)
print(res.choices[0].message.content)
print(res.model_extra["yelno"]) # route, usage and chargeA policy first eliminates every endpoint that violates region, data-retention, or capability requirements — then orders what remains by price, latency, or reliability. Fallback never crosses a hard constraint.
Every model from every provider in the catalog, each with its final price, data policy and gateway-observed health per endpoint.
Browse the model catalog →Compare models →Leaderboards and provider terms →
Every model lists its final price — per token for chat, per unit for media. A request is charged at that price and the exact charge comes back on the response. No subscription, no separate platform fee.
Health is tracked per provider endpoint. Errors are standardized, circuits open per failure class, and eligible traffic re-routes inside your constraints.
SSO and SCIM, roles, approvals, budgets and data policy are enforced server-side and recorded in an audit trail members cannot edit or delete.
OIDC or SAML 2.0 sign-in with enforced domains, SCIM provisioning with group-mapped roles and deprovisioning, and access reviews.
Permitted processing regions, retention ceilings, and no-training rules inherited by every project.
Risk-based confirmation: production routing and price-affecting changes can require quorum approval.
A self-serve IP allow-list on the gateway; PrivateLink and VPC peering are requested in the console and fulfilled manually by platform staff.
Wrap stored provider credentials with a key your organization supplies; external KMS references are registered, not enforced.
OTLP/HTTP delivery of request metadata — metrics and logs — to your collector, with content never included.
The moments that decide whether an AI feature survives production — and what changes when one gateway owns them.