Proxed AI is a secure proxy backend for iOS apps that keeps OpenAI and other LLM API keys behind a server and uses Apple's DeviceCheck to verify requests come from legitimate devices, and it is built for mobile developers who want AI features without shipping credentials inside the app.
What it is
Proxed AI is a TypeScript monorepo that inserts a proxy layer between an iOS application and a large language model provider. The stack is Bun, React, Next.js, Hono, Supabase, shadcn and TailwindCSS, with hosting split across Supabase for auth, database and storage, and Vercel for the landing page, portal and api. Supporting services are Resend, GitHub Actions and Polar. It lives in Apple's iOS ecosystem and in the Next.js and Supabase web ecosystem, and the repository states AGPL-3.0 for non-commercial use.
The concrete problem it solves is the standard shortcut in mobile AI work: an app that calls an LLM directly has to embed the key in the binary or fetch it from a weak endpoint, which exposes the credential, leaves spend uncontrolled and leaves the response shape undefined. Proxed AI replaces that direct client-to-provider call with a request to a proxy that holds the key, checks the device through DeviceCheck, and shapes the response against a schema the developer defined visually. Rate limits and model selection are managed on the server side rather than in the app.
Key capabilities
- Secure key management: OpenAI and other LLM keys stay behind the proxy instead of inside the client.
- DeviceCheck verification: Apple's DeviceCheck integration blocks illegitimate requests at the door.
- Structured responses: a visual schema builder defines and enforces JSON response schemas for AI output.
- Quick integration: the README states setup in under 10 minutes with no external code.
- Observability and control: usage monitoring, rate limits and AI model management happen in real time.
- Monorepo stack: TypeScript, Bun, React, Next.js, Hono, Supabase, shadcn and TailwindCSS in one repository.
- Managed dependencies: Supabase covers auth, database and storage, while Vercel hosts the landing page, portal and api.
Who uses it and how
- iOS teams shipping AI features that need credentials kept out of the distributed binary.
- Apps that need typed JSON back from a model, with the contract defined through the visual schema builder so client-side decoding stays predictable.
- Operators who need per-app rate limits and live usage monitoring to control model spend.
- Self-hosters who already hold Supabase and Vercel accounts, with Resend for email, Polar for billing and GitHub Actions for CI.
- Early adopters working against a young repository: 8 stars, 0 forks and 0 open issues at the time of writing.
Getting started
The README points to documentation that is itself marked as in progress at https://docs.proxed.ai, and the hosted product runs at https://proxed.ai. The README does not name a published npm package, a Docker image or a compose file, so self-hosting currently means building from the monorepo.
How it compares
No list of paid products that this project replaces is supplied in the facts, and no similar tools are named alongside it, so within this registry Proxed AI stands alone. The page therefore offers no direct substitute to compare against, and any evaluation has to run against whatever in-house proxy a team already operates.
When to use it — and when not to
Self-hosting means operating Supabase for auth, database and storage, a Vercel deployment for the landing page, portal and api, and the third-party services Resend for email, Polar for billing and GitHub Actions for CI, which is a real operational load for a small team. The licence is stated as AGPL-3.0 but qualified as being for non-commercial use, a contradiction that commercial adopters should resolve before committing to the code. Documentation is marked as in progress, the repository shows 8 stars and no forks, and DeviceCheck is Apple-only, so teams building for Android, web or desktop should look elsewhere.
