OpenGem – A Load-Balanced Gemini API Proxy (No API Key Required)
Hi HN! I built OpenGem, an open-source, load-balanced proxy for the Gemini API that requires absolutely no paid API keys. GitHub: https://github.com/arifozgun/OpenGem The Context: Like many developers, I was constantly hitting "429 Quota Exceeded" errors while building AI agents and processing large payloads on free tiers. I wanted to build freely without calculating API costs for every test request. How it works: I reverse-engineered the official Gemini CLI authentication to get standard API access. However, a single free Google account quota depletes quickly. To solve…
What it does
In the maker’s words, at launch
Hi HN! I built OpenGem, an open-source, load-balanced proxy for the Gemini API that requires absolutely no paid API keys. GitHub: https://github.com/arifozgun/OpenGem The Context: Like many developers, I was constantly hitting "429 Quota Exceeded" errors while building AI agents and processing large payloads on free tiers. I wanted to build freely without calculating API costs for every test request. How it works: I reverse-engineered the official Gemini CLI authentication to get standard API access. However, a single free Google account quota depletes quickly. To solve this, I built a Smart Load Balancer at the core of OpenGem. What it does: - You connect multiple idle/free Google accounts to the dashboard via OAuth. - OpenGem acts as a standard endpoint (`POST /v1beta/models/{model}`). - It routes traffic to the least-used account. If an account hits a real 429 quota limit, OpenGem instantly detects it, puts that account on a 60-minute cooldown, and seamlessly retries with the next available account. It differentiates between simple RPM bursts and actual limits. Tech specs: - Fully compatible with official Google SDKs (`@google/genai`), LangChain, and standard SSE streaming (no broken [DONE] chunks). - Supports native "tools" (Function Calling) for agentic workflows. - Raised payload limit to 50MB for massive contexts. - AES-256-GCM encryption for all sensitive configs and OAuth tokens at rest. - Toggle between Firebase Firestore or a fully offline Local JSON database. It’s strictly for educational purposes and personal research to bypass the friction of testing/prototyping. The entire project is MIT licensed. I’m currently running it with my own side projects and it handles heavy agent tasks flawlessly. I would love any feedback on the load balancing logic, security implementations, or just general thoughts!
Does the same job
all alternatives →
OpenLIT's Zero-code LLM ObservabilityOct 2025 · ▲136Trace LLM requests + costs with OpenTelemetry monitoring
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
