Proxed.AI

Changelog

Stay updated on Proxed.AI's development journey. Read about our latest features, engineering insights, and product announcements as we build the future of AI model management.

05Jul2025

Google AI (Gemini) Support Now Live!

We're excited to announce that Google AI (Gemini) is now fully supported by Proxed.AI! You can now proxy requests to Google's powerful Gemini models with the same enterprise-grade security and features you've come to expect.

#
What's New:

  • Complete Google AI Integration: Route all your Gemini API calls through Proxed.AI's secure gateway
  • Latest Gemini Models: Support for Gemini 2.5 Pro/Flash, 2.0 series, 1.5 series, and more
  • Unified Security: Same DeviceCheck verification and partial key system for your Google AI keys
  • Structured Responses: Full support for text, vision, and PDF analysis with schema validation
  • Transparent Pricing: Accurate cost tracking based on Google's official pricing tiers

#
Supported Models:

  • Gemini 2.5 Series: gemini-2.5-pro (state-of-the-art), gemini-2.5-flash (best price-performance)
  • Gemini 2.0 Series: gemini-2.0-flash, gemini-2.0-flash-lite
  • Gemini 1.5 Series: gemini-1.5-pro, gemini-1.5-flash, gemini-1.5-flash-8b
  • Specialized Models: Vision models, embedding models, and more

#
API Endpoints:

Route your Google AI requests through:

https://api.proxed.ai/v1/google/{projectId}/{googleApiPath}

For example:

https://api.proxed.ai/v1/google/your-project-id/models/gemini-pro:generateContent

#
Getting Started:

  1. Add your Google AI API key to your Proxed account
  2. Create or update a project to use the Google provider
  3. Use the same authentication methods (Bearer token, DeviceCheck)
  4. Start making secure requests to Gemini models!

This brings our supported providers to three major AI platforms, giving you more flexibility and choice while maintaining consistent security across all your AI integrations.

Check out our documentation and new blog post for implementation details.

06Jun2025

Major Proxy Engine Upgrade: Enhanced Reliability, Security & Observability

We've rolled out a significant overhaul of our core proxy engine, making it more robust, secure, and observable. These updates are now live for all self-hosted instances and are being deployed to our hosted infrastructure.

#
Key Enhancements:

  • Increased Resilience:

    • The proxy now automatically retries failed requests from transient errors (e.g., 429, 502, 503) using an exponential backoff strategy.
    • Requests have configurable timeouts to prevent them from hanging.

  • Advanced Security:

    • We've implemented stricter sanitization of both request and response headers to block sensitive information and prevent potential vulnerabilities.
    • Request validation has been enhanced to protect against issues like HTTP request smuggling.

  • Improved Observability:

    • A new internal metrics collection system tracks request counts, error rates, and latency histograms, providing better insight into performance.

  • Centralized & Future-Proofed Configuration:

    • All AI provider-specific settings are now managed in a centralized, type-safe configuration, making the system easier to maintain and extend.

These foundational improvements enhance the reliability and security of all requests proxied through Proxed.AI and pave the way for our upcoming global multi-region hosting for lower latency worldwide.

For more technical details, see our latest blog post.

08May2025

Anthropic API Proxy Now Live!

Exciting news! Proxed.AI now officially supports Anthropic API proxying, allowing you to securely integrate models like Claude into your iOS apps.

Route your Anthropic calls through Proxed.AI to gain:

  • Top-tier Security: Utilize Apple's DeviceCheck and our Partial Key system for your Anthropic keys.
  • Centralized Monitoring: Track usage, manage costs, and get alerts for both OpenAI and Anthropic in one dashboard.
  • Streamlined Workflow: Enjoy a unified experience for managing multiple AI providers.

We've added a new Swift code example for the Anthropic proxy in our Examples section. This shows how to call the /v1/anthropic/<your-project-id>/messages endpoint.

Explore our documentation for further details.

02May2025

New Feature: Configurable Consumption Notifications

To help you better monitor API usage and manage costs, we're introducing Configurable Consumption Notifications for your projects.

#
Key Features:

  • Custom Thresholds: Define specific API call volume thresholds that trigger an alert.
  • Time Intervals: Set the time window (e.g., 5 minutes, 1 hour, 1 day) over which these calls are counted.
  • Email Alerts: Receive notifications (initially via email) when your project exceeds its configured threshold within the defined interval.
  • Alert Fatigue Prevention: Once a notification is sent, another won't be triggered for the same project until the defined time interval has passed, even if high usage continues.

#
Benefits:

  • Cost Management: Get early warnings of high usage to investigate and prevent unexpected cost escalations.
  • Abuse Detection: Identify potential compromises or misuse indicated by sudden usage spikes.
  • Performance Insights: Gain insights into your app's API consumption patterns.

Configure these notifications in your Project Settings within the Proxed.AI dashboard to define what constitutes high consumption for your specific application needs.

28Mar2025

API Authentication Update: Bearer Token Method Recommended

This update focuses on refining our API authentication mechanisms, with the Bearer Token method now being the primary recommended approach for authenticating your requests to Proxed.AI.

#
Bearer Token Authentication (Authorization: Bearer your-partial-key.your-token):

  • Flexibility: The your-token part can be either a DeviceCheck token (for production) or a Test Key (when Test Mode is enabled for your project).
  • SDK Compatibility: This format is designed for better compatibility with standard AI SDKs that primarily support the Authorization header, allowing both the partial API key and the necessary token to be transmitted in a single, standard header.
  • Security: Combines the security of partial keys with DeviceCheck or the convenience of test keys in a unified manner.

#
Other Supported Methods:

  • Test Key (Header): Using x-proxed-test-key and x-ai-key for development when Test Mode is active.
  • Device Token (Header - Legacy): Using x-device-token and x-ai-key. We encourage migration to the Bearer token method from this legacy approach.

Refer to our API Authentication documentation for full details on each method and best practices.

16Mar2025

Enhanced Developer Experience: New Onboarding & Docs

We've shipped a significantly improved onboarding experience and enhanced our documentation to make getting started with Proxed.AI smoother and more intuitive.

#
Key Improvements:

  • Guided Onboarding Flow: A new linear, step-by-step process guides you through:
    • Quick team creation.
    • Simplified member invitations with role selection.
    • Visual guides for DeviceCheck credential setup.
    • Smarter partial key detection and secure client-side splitting.
  • Simplified DeviceCheck Configuration: Visual explanations, inline documentation for Apple credentials, and better error messaging.
  • Improved Partial Key Management: Auto-detection of key types, one-time secure client-side splitting, and clearer visuals.
  • Contextual Documentation: UI components now link directly to relevant documentation sections (e.g., Partial Keys, DeviceCheck).

These changes are aimed at reducing friction and helping you securely integrate AI into your iOS apps more quickly.

21Feb2025

New API Routes: Text and PDF Analysis with Structured Responses

Expanding our AI analysis capabilities, Proxed.AI now offers new dedicated API routes for Text and PDF analysis, leveraging our Structured Response feature.

#
/v1/text - Text Analysis:

  • Submit text content to this endpoint.
  • Receive structured, schema-validated responses for tasks like:
    • Sentiment analysis
    • Topic extraction
    • Text summarization
    • Key phrase identification

#
/v1/pdf - PDF Analysis:

  • Submit PDF documents (via URL or base64-encoded data).
  • Ideal for:
    • Document summarization
    • Information extraction from PDFs
    • Table and figure analysis

Both routes are fully integrated with the schema you define in your project settings, ensuring predictable and validated outputs. They also benefit from DeviceCheck security, metrics tracking, and comprehensive error handling.

Check our documentation for implementation details.

18Feb2025

Proxed.AI Private Beta Launch!

We are thrilled to announce the Private Beta Launch of Proxed.AI!

Our open-source platform, designed to make securing AI API integrations simple and effective for iOS developers, is now available for you to use. This launch brings together key features developed to address common security and usability challenges:

  • One-URL Security Concept: Easily secure your AI API calls by routing them through Proxed.AI.
  • Core Security Features: Includes robust DeviceCheck integration and our innovative Partial Key management system.
  • Structured Data: Leverage the Visual Schema Builder to define and receive structured, validated responses from AI models.
  • Initial Analytics: Monitor API usage and track performance via the dashboard.
  • Open Source Commitment: Our codebase is on GitHub, and we welcome community involvement.

We're launching with initial support for OpenAI and Anthropic APIs, with more to come.

#
Get Started During Beta:

  • Free Tier: Available for up to 1,000 API calls/month.
  • Pro & Ultimate Tiers: For higher usage needs.
  • Explore the Documentation and GitHub Repository.

This is a significant milestone, and we look forward to your feedback as we continue to evolve Proxed.AI!

13Feb2025

Engineering Update: Secure Token Generation & Key Management Details

This update provides more insight into our multi-layered security approach for API key management and token generation.

#
Core Security Layers:

  1. Split-Key Architecture: API keys are cryptographically split. The server part is stored securely within Proxed.AI infrastructure, while the client part (with metadata including version and a unique split ID) resides on the device. Both are required for key reconstruction, which only happens momentarily during request processing.
  2. Apple DeviceCheck Integration: Every API request requires a valid DeviceCheck token, cryptographically signed by Apple, ensuring requests originate from genuine iOS devices running your application.
  3. Client-Side Key Splitting: The process of splitting keys, generating cryptographic salt, and embedding metadata is designed to happen client-side, further minimizing exposure.

This system ensures that no single location holds a complete API key and that only verified devices can initiate requests, forming the foundation of our zero-trust key management strategy.

12Feb2025

New Feature: Visual Schema Builder for Structured LLM Responses

We're excited to roll out the Visual Schema Builder, a powerful new tool within the Proxed.AI dashboard designed to make working with structured LLM responses more intuitive and efficient.

#
Key Capabilities:

  • Visual Schema Design: Utilize a drag-and-drop interface to construct your desired LLM response schemas.
  • Import & Export: Seamlessly import existing schemas (e.g., from JSON examples or Zod code) and export ready-to-use validation code (Zod schemas, Swift structs).
  • Real-time Validation: Get instant feedback, type checking, and validation for required fields and custom rules as you design.
  • Code Generation: Automatically generate Zod schemas for runtime validation and Swift structs for your iOS client applications.
  • Advanced Features: Support for nested objects, arrays, enums, optional fields, default values, and custom descriptions to guide the LLM.

This tool aims to simplify the process of ensuring your AI model outputs are predictable, reliable, and easily integrable into your applications.

09Feb2025

Introducing the Proxed.AI Core Solution

Today we're detailing the core Proxed.AI solution, designed to offer a simplified and secure way to integrate AI APIs into your iOS applications.

#
Key Architectural Highlights:

  • Secure API Gateway: Proxed.AI acts as a managed (or self-hostable) proxy, meaning your app calls our endpoints, not AI providers directly.
  • Split-Key Management: We employ a split-key approach where provider API keys are divided. A client-side partial key is used by your app, and the complementary server-side part is stored securely by Proxed. Full keys are only momentarily reconstructed during request processing, never fully stored in one place accessible from the client.
  • Built-in DeviceCheck Verification: Every request routed through Proxed.AI undergoes automatic DeviceCheck token validation with Apple's servers, ensuring calls originate from legitimate instances of your app on genuine iOS devices.
  • Additional Safety Features: The platform is built with rate limiting, usage tracking, and cost monitoring capabilities in mind.

Our aim is to provide an easy-to-integrate solution that handles complex security measures like DeviceCheck and secure key management, allowing developers to focus on building features.

29Dec2024

Proxed.AI Project Kick-off: Vision for Secure iOS AI Integration

We're excited to announce the formal start of the Proxed.AI project!

Our core mission is to address the complexities and security concerns developers face when integrating AI models into iOS applications. Proxed.AI is envisioned as an open-source, secure gateway for AI calls on iOS.

#
Initial Focus & Core Concepts:

  • Secure Key Management: Designing a robust system to protect API keys from client-side exposure.
  • DeviceCheck Integration: Leveraging Apple's DeviceCheck to verify genuine iOS devices and prevent unauthorized requests.
  • Structured Responses: Enabling developers to define schemas for predictable and validated AI outputs.
  • Usage Tracking: Providing tools to monitor API usage for cost management and performance optimization.

Proxed.AI will be developed as an open-source platform. You can follow our progress and contribute on GitHub.

This marks the beginning of our efforts to build a comprehensive solution for secure and efficient AI integration on iOS.