Shell wrapper redirects Claude Code’s API calls to DeepSeek V4 Pro ($0.87/M output) instead of Anthropic ($15/M), preserving the full tool loop.
Key Takeaways
Sets ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, and model env vars per-session so Claude Code CLI targets DeepSeek without permanent config changes.
DeepSeek V4 Pro scores 96.4% on LiveCodeBench; auto context caching drops repeat-turn costs to $0.004/M vs $0.44/M uncached.
Supported backends: DeepSeek (default), OpenRouter ($0.44/M input), Fireworks AI ($3.48/M output, US servers), Anthropic fallback via --backend anthropic.
Known limits: no image/vision input, no parallel tool use, no MCP server tools through the compatibility layer.
Remote control mode splits traffic: WebSocket bridge stays on Anthropic infrastructure; model calls route through a local Node.js proxy to DeepSeek.
Hacker News Comment Review
Commenters quickly noted the wrapper is roughly four env var assignments; DeepSeek’s own docs already document Claude Code integration directly, making the added tooling marginal.
Skepticism around model quality: consensus is DeepSeek matches Sonnet-tier for routine tasks but falls short of Opus on complex reasoning, so savings depend on workload mix.
Practical alternatives mentioned include opencode, pi.dev, and Kimi subscriptions as full harness replacements rather than env var shims on the closed Claude Code client.
Notable Comments
@aftbit: posts the functional equivalent as a five-line shell script, implying the repo’s main value is documentation, not code.
@Lihh27: “You’re still betting the whole loop on Anthropic’s closed client” – flags lock-in risk the README does not address.
