Wokey

Wokey is a unified AI API gateway built around an OpenAI-compatible interface. Existing OpenAI SDKs, agents, CLIs and backend services can switch the base URL, use a Wokey platform key and select a model ID without adopting a new request format. The platform handles routing, billing, limits and supply state across models from providers such as OpenAI, Anthropic, DeepSeek, Gemini and others.¹

The more distinctive element is Provider Node. Wokey lets operators connect authorized upstream capacity from their own machines so requests can exit from the provider side rather than from a centralized relay. Wokey says the node supports macOS, Linux, Windows and Docker, maintains a WebSocket link to the platform, and keeps upstream traffic exiting from the provider’s own network. The provider-node project is described as Apache-2.0 open source; that does not make the whole Wokey platform open source.²

Wokey also exposes image-generation/editing endpoints and a browser Studio surface, but the core product reviewed here is the API gateway rather than the consumer image UI.¹

Wokey uses usage-based, model-specific pricing rather than a single subscription price. Its public model catalog says model IDs and prices vary with supply and network conditions and that actual settlement uses the price in effect when the request is processed.³

That structure can be attractive for users who want pay-as-you-go access across several providers without maintaining separate accounts, but it makes cost forecasting less stable than a fixed catalog. The refund policy says completed top-ups, balances, API usage and delivered digital capacity are generally non-refundable except for limited cases such as legal requirements, duplicate charges, clear billing mistakes or qualifying unauthorized charges.⁴

Wokey’s strongest evidence is its own technical transparency rather than independent market validation. The API documentation clearly explains the OpenAI-compatible path, while the provider-node architecture is described in enough detail to understand how request routing differs from conventional centralized relays.¹²

The privacy policy is unusually specific for a young gateway. It says normal successful request tables mainly retain request metadata, token usage and cost rather than full prompts and outputs, but failed, abnormal or investigative requests may have their full original payload archived for up to 14 days. Relevant request content is necessarily sent to the selected upstream model provider. Provider credentials are stored encrypted, with the policy naming AES-256-GCM, and passwords are hashed with Argon2.⁵

Those disclosures improve transparency, but they also matter operationally: organizations sending sensitive prompts should understand that exception payloads can be retained and reviewed by authorized admins, and that traffic can cross jurisdictions depending on the selected upstream provider.⁵

Independent evidence is currently sparse. Major software-review, technology and benchmark sources do not provide a meaningful exact-product benchmark, review corpus or reliability study for Wokey. An Apple App Store result for a shopping app developed by “Shanghai Wokey Technology Co., Ltd” is an unrelated product and should not be treated as evidence about this API gateway. The absence of mature independent coverage limits confidence in production reliability without constituting negative evidence.

Wokey is best suited to developers and small teams already using OpenAI-compatible clients that want a single key across several model providers. It can also appeal to technically sophisticated users who value a provider-side supply network and want access to models without separately wiring each upstream API.¹²

It is less suitable for organizations that require mature enterprise SLAs, static pricing, independent uptime evidence, contractual assurances around upstream policy compliance or strict controls preventing any temporary retention of full failed-request payloads.