elygate

Bifrost is a high-performance AI gateway designed to unify access to over 23 AI providers—including OpenAI, Anthropic, AWS Bedrock, Google Vertex, Azure, and more—through a single OpenAI-compatible API. It eliminates the need for Redis by leveraging Bun and PostgreSQL 18, offering features like semantic caching, O(1) billing, auto-partitioning, and automatic failover. Benchmarks show it adds only 11 µs of overhead per request at 5k RPS, with 100% success rate and sub-microsecond queue wait times. Deployment is simplified: users can start locally via NPX or Docker in under a minute, then configure through an embedded web UI. It supports drop-in SDK replacements for OpenAI, Anthropic, and Google GenAI with one-line URL changes. Advanced capabilities include Model Context Protocol (MCP) for external tool integration, multimodal support (text, images, audio), custom plugins, governance with virtual keys, budget management, OIDC user provisioning, and observability via Prometheus and distributed tracing. The architecture is modular, with components for providers, configuration storage, HTTP transport, UI, and extensible plugins. Enterprise deployments unlock clustering, adaptive load balancing, guardrails, and secrets management. Documentation covers setup, integrations with LangChain, LiteLLM, and others, plus performance benchmarks and contribution guidelines. Licensed under Apache 2.0, built by Maxim.