<p>Claude Code is arguably the best agentic coding tool available today. One catch: it only works with Anthropic’s API. No native support for OpenAI, Bedrock, Vertex AI, or Gemini.</p>
<p><strong>TL;DR:</strong> Bifrost is an open-source Go gateway that translates Anthropic-format requests so Claude Code can route to any LLM provider. One install command, one env var, done.</p>
<p><a href="https://git.new/bifrost" target="_blank" rel="noopener nofollow">Bifrost</a> is an open-source LLM gateway written in Go that fixes this. It exposes an <a href="https://docs.getbifrost.ai/integrations/anthropic-sdk" target="_blank" rel="noopener nofollow">Anthropic-compatible endpoint</a> so Claude Code can talk to any model provider without changing a single line of configuration.</p>
<h2>Why Developers Need This</h2>
<p>For solo work, Anthropic’s API is perfectly fine. But teams hit real friction fast:</p>
<ul>
<li>Compliance rules that mandate a specific cloud provider</li>
<li>Automatic failover when a provider has an outage</li>
<li>Cost optimisation by routing simple tasks to cheaper models</li>
<li>Budget controls across developers and projects</li>
</ul>
<p>Bifrost acts as a translation layer. It accepts Anthropic-format requests via its <a href="https://docs.getbifrost.ai/features/drop-in-replacement" target="_blank" rel="noopener nofollow">drop-in replacement endpoint</a> and forwards them to whatever provider the team configures.</p>
<h2>Setup in 3 Steps</h2>
<h3>1. Install Bifrost</h3>
<p>One command. No Docker, no Kubernetes.</p>
<pre><code>npx -<span class="hljs-attribute">y</span> <span class="hljs-keyword">@maximhq</span>/bifrost
</code></pre>
<p>The gateway starts locally. The <code>/anthropic</code> endpoint is ready to accept requests.</p>
<h3>2. Point Claude Code at Bifrost</h3>
<pre><code><span class="hljs-built_in">export</span> <span class="hljs-attribute">ANTHROPIC_BASE_URL</span>=http://localhost:3000/anthropic
</code></pre>
<p>All Claude Code API calls now route through Bifrost.</p>
<h3>3. Configure the Target Provider</h3>
<p>Set provider credentials in Bifrost’s config. Requests flow through automatically to OpenAI, Bedrock, Vertex AI, Gemini, or any other supported provider.</p>
<p>Claude Code does not know the difference. It sends Anthropic-format requests, Bifrost translates and routes them.</p>
<h2>What Else Bifrost Brings to the Table</h2>
<p>Routing alone would be useful. Bifrost goes further.</p>
<p><strong>Raw speed.</strong> 11 microsecond latency overhead, 50x faster than Python-based alternatives, handling 5000 requests per second. The gateway itself is never the bottleneck. Full <a href="https://docs.getbifrost.ai/benchmarking/getting-started" target="_blank" rel="noopener nofollow">benchmarks are published here</a>.</p>
<p><strong>Automatic failover.</strong> If a provider goes down, Bifrost <a href="https://docs.getbifrost.ai/features/fallbacks" target="_blank" rel="noopener nofollow">reroutes to the next configured provider</a>. No manual intervention. No downtime.</p>
<p><strong>MCP Code Mode.</strong> Reduces token usage by over 50% for coding tasks. Fewer tokens means lower costs and faster responses, which matters in agentic coding where context windows fill up quickly.</p>
<p><strong>Four-tier budget hierarchy.</strong> Rate limiting, cost tracking, and spend caps at the organisation, team, project, and developer level. No custom tooling needed.</p>
<p><strong>Dual-layer semantic caching.</strong> Supports Weaviate, Redis, and Qdrant as vector stores. Repeated similar queries hit cache instead of the provider, cutting costs further.</p>
<h2>Who Should Use This</h2>
<p>Not every developer needs a gateway. If Anthropic’s API works and there is no multi-provider requirement, there is no reason to add one.</p>
<p>Worth noting: Bifrost is self-hosted and relatively new, so you are adding a dependency to your stack.</p>
<p>But for teams that need provider flexibility, failover, or cost governance while using Claude Code, Bifrost is the shortest path. One install command, one environment variable, done.</p>
<p>Source code: <a href="https://git.new/bifrost" target="_blank" rel="noopener nofollow">GitHub</a> | Full setup guide: <a href="https://getmax.im/bifrostdocs" target="_blank" rel="noopener nofollow">Bifrost docs</a></p>

The official daily.dev AI community. Led by thought leaders and AI experts. Stay tuned for insights, news and discussions. 

Bifrost is an open-source Go-based LLM gateway that lets Claude Code route requests to non-Anthropic providers like OpenAI, Bedrock, Vertex AI, and Gemini. It exposes an Anthropic-compatible endpoint, requiring just one install command and one environment variable. Beyond routing, it offers 11-microsecond latency overhead, automatic failover, MCP Code Mode for 50%+ token reduction, four-tier budget controls, and dual-layer semantic caching with Weaviate, Redis, or Qdrant. Primarily useful for teams with compliance requirements, multi-provider needs, or cost governance goals.

How to Connect Non-Anthropic Models to Claude Code with a Gateway