bifrost

Bifrost is a high-performance AI gateway that exposes a single OpenAI-compatible API while routing requests to multiple AI providers under the hood. It also includes an MCP (Model Context Protocol) layer that lets models use external tools and resources (such as filesystems, web search, and databases) to augment capabilities. With zero-config startup and a web UI for configuration, you can enable automatic failover, load balancing, and semantic caching for more reliable and cost-efficient AI workloads. To get started, install and run Bifrost via NPX or Docker, then open the built-in web interface to configure providers, MCP tools, and governance settings. You can send requests to the unified API endpoint and optionally enable MCP-enabled tool integrations to give your models access to external context during conversations. The gateway also supports streaming, multimodal inputs, and enterprise features like observability and security controls.

Prerequisites:

• Node.js and npm installed (for NPX-based setup) or Docker installed (for containerized setup).
• Optional: access to provider keys (OpenAI, Anthropic, AWS Bedrock, etc.) depending on your deployment.

Installation options:

1. NPX (quick start, local development):

# Install and run Bifrost gateway in one command
npx -y @maximhq/bifrost

2. Docker (production-ready):

# Run the gateway using Docker and expose port 8080
docker run -p 8080:8080 maximhq/bifrost

3. Web UI configuration after startup:

• Open http://localhost:8080 in your browser.
• Configure providers (OpenAI, Anthropic, AWS Bedrock, Vertex AI, etc.), MCP tool integrations, caching, and governance as needed.
• Use the MCP features to enable external tools for models, such as web search or file access, within your API calls.

Tips and caveats:

• NPX command uses the @maximhq/bifrost package; ensure network access to fetch the package when running for the first time.
• If running via Docker, ensure the container has access to any required provider credentials and configuration files mounted as needed.
• MCP features enable models to use external tools; plan access controls and caching policies to balance cost and latency.
• The gateway offers automatic failover and load balancing; configure multiple providers or keys to leverage these capabilities.
• Monitor observability and logs via the built-in UI or your preferred Prometheus/Tracing setup for production deployments.