Bifrost (US) is an open-source AI gateway written in Go and maintained by Maxim AI. It sits between the application and model providers, focused on low latency, adaptive load balancing and scale on your own infrastructure.
What it does
- Routing and failover across providers such as OpenAI, Anthropic, Google, Bedrock and local models
- Adaptive load balancing across keys and deployments
- Guardrails and access policies enforced at the gateway
- Observability, metrics and cost control
- Cluster mode for high concurrency
How it works
- The application points at an OpenAI-compatible endpoint
- The gateway picks the active route and retries the call on failure
- Configuration lives in a file, with an optional admin panel
- It installs on your own infrastructure, as a container or binary
Models and providers
- Dozens of providers and hundreds of models, including open options
- Text, image, audio and embedding models
- Providers connected with your own key (BYOK)
Public pricing
- Open-source core free under the Apache 2.0 license
- Enterprise edition under quote, with support and governance features
- No mandatory managed service
Strengths
- Low per-call overhead, built for high volume
- Self-hosted deployment with no external service dependency
- Failover and balancing without changing application code
Points of attention
- Requires your own infrastructure operations and monitoring
- Full governance features are in the paid edition
- Route selection depends on correct key configuration