bifrost
Bifrost is a high-performance AI gateway that unifies access to 23+ providers (OpenAI, Anthropic, AWS Bedrock, Google Vertex, Azure, Cerebras, Cohere, Mistral, Ollama, Groq and others) through a single OpenAI-compatible API. You start it with `npx -y @maximhq/bifrost` or `docker run -p 8080:8080 maximhq/bifrost`, configure providers in the built-in web UI at localhost:8080, and call `/v1/chat/completions` as usual.
The README groups its features into core infrastructure (unified interface, multi-provider support, automatic fallbacks between providers and models, load balancing across keys and providers), advanced features (MCP for tool use, semantic caching, multimodal and streaming support, custom plugins, governance with usage tracking, rate limits and access control), enterprise and security (hierarchical budgets with virtual keys, teams and customers, Google and GitHub SSO, Prometheus metrics, distributed tracing, secrets management) and developer experience (zero-config startup, drop-in replacement for the OpenAI, Anthropic and GenAI APIs, SDK integrations, configuration through web UI, API or files). A Go SDK is available for direct integration, enterprise private deployments are offered, and the code is Apache-2.0 licensed.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-25
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 30
- Latest release
- transports/v2.3.2 (2026-10-20)
- Open issues
- 1336
- Recent issues with at least one comment
- 1/7
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Who it fits
- You operate high-throughput production AI applications and need an ultra-fast LLM gateway (50x faster than LiteLLM, <100 µs overhead at 5k RPS).
- You need unified access across 15+ model providers with automatic failover, adaptive load balancing, cluster mode, and semantic caching.
- You are running small low-traffic pet projects where direct provider SDK calls are sufficient without dedicated proxy infrastructure.
- You need visual prompt chaining, vector RAG pipelines, or chatbot web interfaces (Bifrost is an infra-level API router).
How it differs
vs Flowise★ 55.5kBifrost operates at the network gateway layer with sub-millisecond failover; Flowise sits at the application orchestration layer for dragging agent workflows.
vs agentgateway★ 5.7kBifrost specializes in raw LLM token throughput, failover, and caching; AgentGateway focuses on agent identity, security authorization, and tool governance.
Compatibility
Quick start
Use cases
Alternatives
Comments
No comments yet. Be the first!