MCP & Tool Calling

Bifrost AI Gateway

maximhq/bifrost

Bifrost is a Go-based AI gateway that provides a unified, OpenAI-compatible API for 23+ AI providers. It supports routing, fallbacks, load balancing, semantic caching, governance, observability, multimodal requests, plugins, and Model Context Protocol integrations.

★ 7KStars
⑂ 984Forks
778Open issues
GoLanguage
Apache-2.0License
Q@project.QualityScoreEditorial score

Overview

The project can run as an HTTP gateway, be embedded through its Go SDK, or serve as a replacement endpoint for supported provider SDKs. A built-in web interface supports configuration, monitoring, and analytics. The repository describes both open-source capabilities and additional enterprise deployment features.

Key features

  • Single OpenAI-compatible interface for 23+ providers
  • Automatic retries, provider or model fallbacks, and load balancing
  • Text, image, audio, and streaming request support
  • Semantic response caching based on similarity
  • MCP gateway capabilities for external tool access
  • Virtual keys, usage tracking, budgets, rate limits, and fine-grained access control
  • Prometheus metrics, distributed tracing, logging, monitoring, and analytics
  • Web, API-driven, and file-based configuration
  • Extensible plugin architecture
  • HTTP gateway, Go SDK, and replacement endpoints for supported SDKs
  • Enterprise options described for clustering, adaptive load balancing, guardrails, private networking, OIDC provisioning, custom controls, and governance

Requirements, installation and quick start

NPX quick start: `npx -y @maximhq/bifrost`. Basic Docker start: `docker run -p 8080:8080 maximhq/bifrost`. Docker with persistent data: `docker run -p 8080:8080 -v $(pwd)/data:/app/data maximhq/bifrost`. Go SDK installation: `go get github.com/maximhq/bifrost/core`.

Usage

After starting the gateway, open `http://localhost:8080` to configure providers through the web interface. A chat request can be sent with `curl -X POST http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"openai/gpt-4o-mini","messages":[{"role":"user","content":"Hello, Bifrost!"}]}'`. Existing OpenAI, Anthropic, or Google GenAI integrations can instead point their base URL or API endpoint at the documented local Bifrost endpoints.

Model compatibility and use cases

The README states support for 1,000+ models across 23+ providers. Named providers include OpenAI, Anthropic, AWS Bedrock, Google Vertex, Azure, Cerebras, Cohere, Mistral, Ollama, and Groq. It also documents integrations for OpenAI, Anthropic, AWS Bedrock, Google GenAI, LiteLLM, and LangChain SDKs. A complete model list is not included in the supplied repository record.

License and risk notes

The repository metadata identifies the license as Apache-2.0, and the README states that the project is licensed under the Apache 2.0 License. Review the repository LICENSE file before redistribution or deployment.

Release and maintenance

Not stated in the repository metadata.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33.9KPython

Composio

composiohq/composio

★ 29.5KTypeScript