OmniaKey

OmniaKey 是什么？

Category: a unified AI API platform / gateway — one prepaid account and one API key in front of frontier text models and asynchronous media models. The problem it solves Teams building with AI today end up with a wallet, an invoice, an SDK quirk and a rate limit at every provider. Coding agents make it worse: Claude Code speaks Anthropic's protocol, Codex speaks OpenAI's Responses API, Gemini SDKs speak their own, and each expects a different base URL and credential. Image and video models add a second, unrelated integration: submit → poll → download, with per-provider quirks. OmniaKey collapses that into one host, one key, one balance, one usage ledger. 1. One key, three native protocol surfaces Everything lives behind https://api.omniakey.com Protocol Base URL Typical clients OpenAI-compatible /v1 OpenAI SDK, Cursor, Codex, Cline, aider Anthropic native bare host (/v1/messages appended by the SDK) Claude Code, Anthropic SDK Gemini native /v1beta Google Gen AI SDK No proprietary SDK is required — you point an existing client at a new base URL and keep your code. Beyond Chat Completions it exposes OpenAI Responses, Anthropic Messages, Gemini generateContent, model-discovery endpoints for both OpenAI and Gemini shapes, and a typed decision endpoint (/v1/alpha/search). In practice: export ANTHROPIC_BASE_URL=https://api.omniakey.com plus a token, and Claude Code is running against the whole catalog. 2. A broad, honestly-labelled catalog Roughly 60 published model IDs across 12 providers: Anthropic (Claude Fable / Opus / Sonnet / Haiku), OpenAI (GPT‑6 and 5.6 families), Google (Gemini Pro / Flash / Flash‑Lite), xAI (Grok), Alibaba (Qwen), DeepSeek, Z.ai (GLM), Moonshot (Kimi), MiniMax, Xiaomi (MiMo), Tencent (Hunyuan) and TypeSafe (Jev). Most carry 1M context, vision and tool use; low-latency fast variants are marked as such. Two deliberate policies sit on top of the catalog: No silent substitution. The model ID you send is the model that runs — no quantizing, distilling or swapping to a cheaper route behind your back. No silent failover. If a provider is down you get an error you can see, not a different model answering your prompt. 3. Media as first-class async tasks Image, video and audio share one lifecycle instead of three integrations: POST /v1/media/tasks → poll GET /v1/media/tasks/{id} → download from a signed URL (or cancel). Idempotency-Key is required, so a retried submit returns the original task instead of double-charging you; a reused key with a different body returns 409. Statuses are normalized (queued / processing / succeeded / failed / canceled), with progress percent and queue position. Billing is pre-authorized at creation and refunded automatically if the task ends failed or canceled. Results are short-lived signed URLs (~1 hour); re-polling the task re-signs them. Task records persist 24 hours. Live today: GPT Image / Nano Banana class image generation and editing, and Grok Imagine Video at 720p billed per output second. The audio route is documented on the same protocol, but no public audio model is enabled yet — creation fails rather than pretending. 4. Pricing that stays boring Prepaid USD balance, metered by actual usage: per token for text (input / output / cache rates published per 1M tokens), per image and per output second for media. No subscription, no tiers, no seat count. Selected models carry launch discounts against official list price, shown per model rather than as one blended claim. Each API key can also carry its own spend cap, which is what makes handing a key to a teammate or an agent safe. Top-up via Stripe (cards, Apple Pay, Google Pay), Creem, or NOWPayments for crypto. Invoicing is available on request. 5. The operating surface Dashboard — key CRUD, balance, top-ups, spend-by-day, model mix, latency and a full call ledger with per-call cost. Developer tools — a CLI, a local stdio MCP server and an Agent Skill from one package (omniakey-com/omniakey-devtools), pinned by source tag, key read only from OMNIAKEY_API_KEY. Submit once, resume by task ID; timeouts never auto-create a second paid job. Web generators — image, video, audio and text workbenches that use your session rather than minting a key. Changelog & docs — model additions, repricing and status in one feed; docs in 12 languages with runnable examples for Claude Code, Cursor, Codex, Cline, aider and the OpenAI SDK. Privacy posture — prompt and response bodies are not stored by default; request metadata (model, tokens, latency, cost) is kept for billing and the usage view. 6. Explicit scope boundaries OmniaKey is a gateway, not a model host: no embeddings, rerank, files, fine-tuning or batch APIs today. It is also not a subscription reseller — it does not replace a Cursor Pro seat or an Anthropic plan, it replaces the credentials and billing behind them.

OmniaKey 速览

域名评级

DR 43Domain Rating by Ahrefs许可协议