1. Open-Source Infrastructure Forever
We believe that the critical control layer connecting your AI applications to foundation models must remain open, auditable, and self-hostable. Infrastructure vendor lock-in is a risk to software independence.
Mirrabase is published under the MIT license. Every core capability—including unified model proxying, failover routing, and PII redaction—is developed transparently in the open. You can inspect the source code, deploy it in air-gapped environments, or modify it to fit your internal compliance requirements without restrictions.
2. Sub-5ms Gateway Overhead SLA
Adding a security and governance proxy should never degrade user experience or introduce latency bottlenecks. Mirrabase’s core routing engine is written in Rust to guarantee sub-5ms overhead latency on streaming LLM responses.
Our routing engine continuously evaluates upstream provider health. If OpenAI, Anthropic, or DeepSeek experiences elevated latency or outages, Mirrabase instantly switches traffic to your designated fallback model in real-time without dropping active client connections.
3. Zero Telemetry & Complete Data Privacy
Your prompts, completions, agent tool outputs, and customer data belong exclusively to you. Mirrabase operates with zero telemetry collection. We do not store, sample, or train on any prompt data passing through your self-hosted instance.
All audit logs, traces, and metrics remain strictly within your infrastructure boundaries or your private OpenTelemetry data collector.