ghuntley/underclass
Underclass is an OpenAI-compatible pooling proxy written in Rust that optimizes developer subscription API usage. The tool keeps prompt caches warm by pinning sessions to single accounts, automatically handles rate limits by cooling exhausted subscriptions, and returns immediate failure responses based on retry headers when pools are dry.
Underclass functions as a proxy server built in Rust. It manages Open AI compatible requests by routing them through multiple accounts. The system keeps prompt caches active by keeping sessions fixed on specific accounts. It also manages capacity by temporarily suspending subscriptions that have hit their limits. Developers can route application traffic through this proxy to access shared subscription resources. The system identifies when a specific account has no remaining quota during a period. It then shifts new requests to other available accounts in the pool. If all accounts are exhausted, the software returns error headers indicating when capacity might return. The project has accumulated 152 stars and 9 forks on GitHub. This suggests a level of community interest within the open source ecosystem. The tool is described as an Open AI compatible load balancer. Users should verify that their specific API requirements align with Rust based infrastructure. The text notes the project crossed a discovery threshold today, marking its recent arrival in public view.
README
ghuntley/underclass View on GitHub
Loading the README from GitHub…