Everything the gateway does, in one place.

nicrron/auto picks per request; nicrron/reasoning, nicrron/coding and nicrron/fastest pin the job. DeepSeek V4 Pro and GPT-6 Sol for reasoning, Claude Opus 5.5 and Kimi K2.7 Code for code.

EWMA latency per provider, breakers on degraded endpoints, and ordered fallback arrays that cascade on 429 or 5xx before the first byte.

Deterministic requests are served from Redis in about 12 ms at no cost, scoped to your account and keyed by the full request.

Upstream cache reads pass through: 90% off with Anthropic, 50% off with OpenAI, recorded in usage.

OpenAI text-embedding-3, Mistral mistral-embed and codestral-embed, and the Voyage 4 family behind /v1/embeddings with fallback chains.

Seedance 2.0 with native audio, Veo 3.1, Kling and Hailuo for video, Nano Banana for images. Async jobs, exact cost charged upfront, refunded on failure.

OpenAI-compatible TTS from $15 per million characters with six-plus voices, and Whisper at $0.006 per minute billed on true audio duration.

Pooled credits per organization with Owner, Admin, Developer and Viewer roles and invitations.

HMAC-SHA256 signed payloads to any HTTPS endpoint when balance crosses a threshold or daily spend spikes.

Filter live traffic by date, provider, status and cost. Export RFC 4180 CSV or structured JSON in one click.

Run two or three models on the same prompt and compare time to first token, tokens per second and cost, with a blind mode.

Chat, embedding and audio payloads stream through memory only. Nothing is written to disk or a database; only token counts and hashes are kept.