Aurora: AI gateway fork for multi-IP setups, 55x faster than LiteLLM

Description and use

The upgraded Aurora fork addresses OpenCode API session header issues, supports multi‑IP load balancing, offers a web UI for runtime management, and delivers up to 55× performance gains over LiteLLM.

Problem

The original Aurora LLM gateway lacked production features, advanced routing, and a flexible management interface, and recent OpenCode API updates broke local coding tools with HTTP 400 errors.

Outcome

The upgraded Aurora gateway resolves OpenCode API session header issues, supports multi-IP load balancing, provides a web UI for runtime management, and delivers significant performance gains (up to 55x faster than LiteLLM).

Workflow

• 1. Clone the forked repository from https://github.com/entitybtw/aurora.
• 2. Build the Docker image (entbtw/aurora:latest) or compile from source.
• 3. Configure providers and bind_ip settings in the web UI or config files.
• 4. Start the gateway; the Session Hub maps local client traffic to stable session IDs, preventing HTTP 400 errors.
• 5. Use the administrative dashboard to manage routing strategies, fallback chains, and monitor performance.
• 6. Deploy in production with multi-IP load balancing and monitor prompt cache warmness.

Evidence

Fixed OpenCode API: Automatically maps local client traffic into stable, valid session IDs via Session Hub (map mode) to prevent HTTP 400 drops and keep upstream Prompt Cache warm.

Source

Performance: Pure Go architecture, running up to 55x faster than LiteLLM with a lock-free session hot path (~495 ns/op).

Source