whysooraj's Portfolio - geminiPortfolio showcasing projects and research.Zola2024-07-04T00:00:00+00:00https://blog.sooraj.fun/tags/gemini/atom.xmlHow I Got Claude Code Running on My Gemini Subscription (Without Any API Keys)2024-07-04T00:00:00+00:002024-07-04T00:00:00+00:00
Unknown
https://blog.sooraj.fun/blog/codeweaver-launch/<p><strong>TL;DR:</strong> I built a lightweight local proxy called <a href="https://github.com/whysooraj/CodeWeaver">CodeWeaver</a> that translates Claude Code's Anthropic API calls into Google Cloud Code Assist requests — using your existing Gemini CLI OAuth session. Zero dependencies, one file, ~500 lines.</p>
<hr />
<h2 id="the-problem">The Problem</h2>
<p>If you're a developer working with AI coding agents, you've probably noticed a split in the ecosystem:</p>
<ul>
<li><strong>Claude Code</strong> has arguably the best UX among AI coding CLIs — great streaming, tool use, multi-turn conversations, and a polished workflow.</li>
<li><strong>Gemini CLI</strong> comes bundled with your Google account and doesn't require managing API keys, but the CLI experience is different.</li>
</ul>
<p>I found myself wanting the Claude Code interface with my Gemini subscription. Paying for two AI subscriptions felt wasteful, and setting up API keys for every tool was a pain.</p>
<p>So I built a bridge.</p>
<h2 id="the-architecture">The Architecture</h2>
<p>CodeWeaver is a local HTTP proxy that sits between Claude Code and Google's APIs:</p>
<pre><code>Claude Code ──> Anthropic API request ──> CodeWeaver ──> Gemini Request ──> Google
│
Claude Code <── Anthropic response <── CodeWeaver <── Gemini Response <────────┘
</code></pre>
<p>It's a single Node.js file using only native modules (<code>http</code>, <code>fs</code>, <code>path</code>, <code>crypto</code>) — zero npm dependencies. It runs on localhost:8099 and speaks the Anthropic Messages API on one side and the Google Cloud Code Assist streaming API on the other.</p>
<h2 id="how-protocol-translation-works">How Protocol Translation Works</h2>
<p>This was the most interesting part. Claude Code and Gemini have different API shapes, and bridging them required handling several translation layers:</p>
<h3 id="1-message-role-mapping">1. Message Role Mapping</h3>
<p>Anthropic uses <code>"user"</code> / <code>"assistant"</code> roles, while Gemini uses <code>"user"</code> / <code>"model"</code>. Simple enough. The tricky part was handling tool interactions.</p>
<h3 id="2-tool-use-function-calling">2. Tool Use ↔ Function Calling</h3>
<p>When Claude Code calls a tool, it sends a <code>tool_use</code> content block. CodeWeaver translates this into Gemini's <code>functionCall</code> format:</p>
<pre><code>Anthropic: Gemini:
tool_use block ──> functionCall { name, args }
tool_result ──> functionResponse { name, response }
</code></pre>
<p>The trickiest bit was preserving the <strong>thought signature</strong> — Gemini 3 models include a cryptographic signature in tool responses for security. CodeWeaver encodes this inside the tool_use ID with a custom encoding scheme (<code>toolu_{name}_{random}_sig_{sig}</code>) and decodes it on the way back.</p>
<h3 id="3-schema-sanitization">3. Schema Sanitization</h3>
<p>This was the biggest gotcha. Claude Code sends OpenAPI 3.1-style tool schemas with keywords like <code>$schema</code>, <code>anyOf</code>, <code>oneOf</code>, <code>const</code>, and <code>exclusiveMinimum</code>. Google's API strictly rejects these.</p>
<p>CodeWeaver's <code>sanitizeSchema()</code> function strips unsupported keywords, flattens <code>anyOf</code>/<code>oneOf</code> into their typed variant, and converts <code>const</code> to <code>enum</code> arrays — all while preserving the actual type structure.</p>
<h3 id="4-streaming-response-translation">4. Streaming Response Translation</h3>
<p>Claude Code expects Server-Sent Events (SSE) with a specific event structure:</p>
<pre><code>event: message_start
event: content_block_start
event: content_block_delta (text_delta / input_json_delta)
event: content_block_stop
event: message_delta
event: message_stop
</code></pre>
<p>Google sends a different SSE format. CodeWeaver maintains state about the current active content block (text vs. tool_use) and translates Google's stream chunks into the Anthropic event sequence on the fly.</p>
<h2 id="authentication-the-best-part">Authentication: The Best Part</h2>
<p>No API keys. No <code>.env</code> files. No vertex AI setup.</p>
<p>CodeWeaver reuses the OAuth credentials from your existing Gemini CLI session. When you run <code>codeweaver --login</code>, it:</p>
<ol>
<li>Spins up a local callback server on port 8085</li>
<li>Opens a Google OAuth consent screen (PKCE flow)</li>
<li>Exchanges the authorization code for access + refresh tokens</li>
<li>Optionally provisions a free-tier Cloud Code Assist project</li>
<li>Stores credentials locally</li>
</ol>
<p>After that, it handles token refresh transparently — keeping your session alive without any manual intervention.</p>
<h2 id="rate-limiting-and-fallbacks">Rate Limiting and Fallbacks</h2>
<p>Google's Cloud Code Assist has strict per-minute quotas on Pro models. CodeWeaver handles this with a self-healing fallback:</p>
<ul>
<li>If a Pro model (<code>gemini-2.5-pro</code>) hits a 429, the proxy transparently retries with <code>gemini-2.5-flash</code></li>
<li>If the Flash model also hits limits, it parses the retry-after duration and waits</li>
<li>Up to 3 retry attempts before giving up</li>
</ul>
<p>This means your Claude Code session rarely fails — it just gracefully degrades to a faster (but still capable) model when traffic is high.</p>
<h2 id="model-mapping">Model Mapping</h2>
<p>Claude Code has a hard-coded model selector. CodeWeaver maps those to their Gemini equivalents:</p>
<table><thead><tr><th>Claude Code Model</th><th>Gemini Model</th></tr></thead><tbody>
<tr><td>Sonnet (default)</td><td>gemini-2.5-pro</td></tr>
<tr><td>Haiku</td><td>gemini-2.5-flash</td></tr>
<tr><td>Opus</td><td>gemini-3.1-pro-preview</td></tr>
<tr><td>Fable</td><td>gemini-3.1-flash-lite</td></tr>
</tbody></table>
<p>You can also pass raw Gemini model IDs directly: <code>claude --model gemini-3.1-pro-preview</code>.</p>
<h2 id="the-result-500-lines-zero-dependencies">The Result: 500 Lines, Zero Dependencies</h2>
<p>The entire proxy is a single <code>proxy.js</code> file. Here's the breakdown:</p>
<ul>
<li><strong>~80 lines</strong> — OAuth login flow with PKCE</li>
<li><strong>~60 lines</strong> — Schema sanitization</li>
<li><strong>~120 lines</strong> — Request translation (Anthropic → Gemini)</li>
<li><strong>~200 lines</strong> — HTTP server + streaming response handling</li>
<li><strong>~40 lines</strong> — Token management and config</li>
</ul>
<p>Total: ~500 lines of vanilla Node.js. No TypeScript, no frameworks, no npm packages.</p>
<h2 id="what-s-next">What's Next</h2>
<p>CodeWeaver is intentionally focused on the Gemini → Claude bridge right now. But the roadmap aims for something bigger:</p>
<ul>
<li><strong>Plugin architecture</strong> — provider-agnostic backend swapping</li>
<li><strong>Ollama support</strong> — run Claude Code against local models</li>
<li><strong>OpenAI / DeepSeek / OpenRouter</strong> — any API-compatible provider</li>
<li><strong>Auto failover</strong> — cascade through providers on rate limits or outages</li>
<li><strong>Docker image</strong> — containerized deployment</li>
<li><strong>Web dashboard</strong> — visual provider configuration</li>
</ul>
<p>The vision is to make Claude Code a universal frontend for any AI model backend.</p>
<h2 id="try-it">Try It</h2>
<pre data-lang="bash" class="language-bash "><code class="language-bash" data-lang="bash">curl -fsSL https://raw.githubusercontent.com/whysooraj/CodeWeaver/main/claude-gemini-proxy/install.sh | bash
codeweaver --login
codeweaver
claude
</code></pre>
<p>Or clone and run manually:</p>
<pre data-lang="bash" class="language-bash "><code class="language-bash" data-lang="bash">git clone https://github.com/whysooraj/CodeWeaver.git
cd CodeWeaver/claude-gemini-proxy
node proxy.js --login
node proxy.js
claude
</code></pre>
<p><strong>Requirements:</strong> Node.js 18+, a Google account, and the Gemini CLI.</p>
<hr />
<p><em>CodeWeaver is open source under MIT. Contributions, issues, and feature requests are welcome!</em></p>
<p><strong>Links:</strong></p>
<ul>
<li>GitHub: <a href="https://github.com/whysooraj/CodeWeaver">github.com/whysooraj/CodeWeaver</a></li>
<li>Install: One-liner above or clone the repo</li>
</ul>
<hr />
<p><em>Have questions? Found a bug? Want another provider supported? Let me know in the comments or open an issue on GitHub.</em></p>