Proxed AI is an open-source proxy that sits between an iOS app and its AI providers, keeping raw provider keys off the device. It checks each request with Apple DeviceCheck, splits API keys so only a harmless fragment ships in the app, and adds rate limits, spending ceilings, and usage monitoring for iOS developers and small teams shipping AI features without running a backend.[1][2][3]
Quick verdict
iOS teams shipping AI features without a backend will get the most from Proxed AI. Swapping the provider URL for a Proxed endpoint adds DeviceCheck attestation, split-key protection, and per-project spend guardrails with no SDK to install. The constraint is focus and flux: it is an iOS-centered proxy from a small vendor that has paused new cloud registrations while it changes direction, so teams needing many client platforms or a fully open self-serve cloud should look at a larger gateway instead. If you build for iOS and can self-host or already hold a cloud account, start with the free open-source build; if you need a general multi-platform gateway, evaluate LiteLLM instead.[1][2][3]
Key features
Apple DeviceCheck attestation on every request The documented iOS flow fetches a device token and sends it in an x-device-token header, so hardware-level verification blocks emulators, scripts, and jailbroken devices before they reach quota.[1][2]
Split-key protection for provider credentials Only a harmless key fragment ships inside the app while the sensitive half stays on the proxy server, directly answering the hard-coded-key extraction problem.[1]
One-line endpoint swap with no SDK Integration means changing the API URL and headers, with Swift sample code for vision analysis, so adoption rarely means rewriting application code or adding dependencies.[1]
Three provider proxies plus structured endpoints Documented proxies cover OpenAI, Anthropic, and Google AI model families, alongside vision, text, PDF, and image-generation analysis APIs.[1][2]
Visual schema builder for structured outputs Teams define JSON response schemas visually and the proxy enforces consistent, type-safe model output for downstream parsing.[1][2]
Per-project rate limits and spending ceilings Granular guardrails cap usage per project with alerts before budgets blow, addressing the runaway-bill failure mode of a leaked key.[1][2]
Built-in observability and audit trails Usage monitoring, lightweight metrics and health endpoints, and compliance-oriented audit trails expose what each project spends and calls.[1]
Free self-hosting on standard services The open-source codebase self-hosts with Supabase, Vercel, and GitHub accounts and no usage limits, which suits privacy-focused teams and regulated deployments.[1][2]
Who will love it
iOS teams shipping AI features without a backend The endpoint swap plus DeviceCheck gives hardware-backed key protection quickly, with no server code to write or operate.[1][2]
Indie developers worried about key theft and bills Split keys plus per-project ceilings convert an existential leaked-key incident into a capped, alerted event.[1]
Privacy-focused teams that self-host everything The open codebase on your own Supabase and Vercel accounts keeps prompts, keys, and logs inside infrastructure you control.[1][2]
Who should avoid it
Teams shipping beyond Apple platforms DeviceCheck attestation is iOS-specific, so Android, web, or desktop clients sit outside the core protection story.[1][2]
Buyers needing open self-serve cloud onboarding New cloud registrations are paused during the vendor's direction change, so only existing accounts and self-hosting are live paths today.[1]
Enterprises needing broad coverage and attestations Three provider families, email-led support outside the Enterprise tier, and unstated data regions fit small teams better than compliance-heavy procurement.[1][2]
Alternatives and compatibility
LiteLLM for general multi-provider gateway needs LiteLLM is the established open-source AI gateway unifying 100+ models behind one API with budgets and guardrails; prefer it when breadth across providers matters more than iOS attestation.[1]
Portkey AI Gateway for fallbacks and reliability Portkey AI Gateway routes across many providers with automatic fallbacks and observability; prefer it when provider failover is the deciding requirement.[1]
Questions worth asking
What is the best Proxed AI alternative for non-technical users?
Is Proxed AI really no-code?
How much does Proxed AI cost per month?
Sources and editorial review
Source research. Reviewed on October 12, 2026. Written by Sanjib Sah. AI assisted; reviewed by a human editor.
We have not run or tested this product.
- Proxed.AI homepage
official · Observed October 12, 2026
Official homepage: positioning, DeviceCheck and split-key protection, provider proxies, cloud pricing tiers, free self-hosting, paused registrations with existing customers continuing.
- About Proxed
official · Observed October 12, 2026
About page: mission, open-source-first positioning, founder Alex Vakhitov; links to docs, example apps, privacy, terms, status.
- Proxed.AI docs overview
official · Observed October 12, 2026
Official docs: supported providers and models, DeviceCheck, partial keys, structured responses, rate limiting, monitoring, audit trails, streaming, retries, self-hosting requirements, API references.
- github.com nech-ai/proxed
official · Observed October 12, 2026
Repository page: identity, AGPL-3.0 license, 8 stars, 0 forks, 357 commits, official-site link.
- nech-ai/proxed README
official · Observed October 12, 2026
Repository README: features, architecture stack, Supabase and Vercel hosting, AGPL-3.0 for non-commercial use.
- Proxed AI discovery listing
published measurement · Observed October 12, 2026
Discovery lead: official-site and repository identity links, 8 stars, recent commit activity, repository age, license, tags.
Compared with these products
No explicit proprietary baseline is recorded.