deepclaude VS clawcodex

[ericleepi314]

While deepclaude could route the claude code to DeepSeek model, I found it very cumbersome to do so. There are Claude Code rebuilds that natively supports DeepSeek v4 pro. For example I found this one:

https://github.com/agentforce314/clawcodex

The config is easy to set. After that, you can switch model just by slash command /

Here are the config example in their README.md:

{
"default_provider": "anthropic",
"providers": {
"anthropic": {
"api_key": "base64-encoded-key",
"base_url": "https://api.anthropic.com",
"default_model": "claude-sonnet-4-6"
},
"openai": {
"api_key": "base64-encoded-key",
"base_url": "https://api.openai.com/v1",
"default_model": "gpt-5.4"
},
"glm": {
"api_key": "base64-encoded-key",
"base_url": "https://open.bigmodel.cn/api/paas/v4",
"default_model": "zai/glm-5"
},
"minimax": {
"api_key": "base64-encoded-key",
"base_url": "https://api.minimaxi.com/anthropic",
"default_model": "MiniMax-M2.7"
},
"openrouter": {
"api_key": "base64-encoded-key",
"base_url": "https://openrouter.ai/api/v1",
"default_model": "deepseek/deepseek-v4-pro"
},
"deepseek": {
"api_key": "base64-encoded-key",
"base_url": "https://api.deepseek.com",
"default_model": "deepseek-v4-pro"
}
}
}

[aattaran]

clawcodex looks like a Claude Code rebuild (a from-scratch CLI that mimics the surface). deepclaude is a proxy — it works with the official, signed Claude Code binary unmodified.

Trade-offs:

	deepclaude	clawcodex (and other rebuilds)
Claude Code binary	Official, unmodified	Their fork/rebuild
Trust surface	Anthropic + small proxy	Whoever maintains the rebuild
Multi-backend switching	Yes (mid-session via /deepseek etc.)	Yes (their slash commands)
Subscription support	Yes (Claude Max OAuth flows through, PR #26)	Depends on rebuild
Setup	Wrapper + proxy on :3200	Their config file
Updates from upstream Claude Code	Automatic (you run real claude)	Only when their fork pulls them

If you'd rather not bet on a rebuild keeping pace with the official Claude Code, deepclaude lets you stay on the official binary and add backend flexibility around it. If you want a single config file and don't mind running a rebuild, clawcodex is fine. Different bets.

[ericleepi314]

Thanks for the breakdown @aattaran — fair framing of the basic trade-off, but I think a few rows oversimplify in deepclaude's favor:

Trust surface. "Anthropic + small proxy" vs "whoever maintains the rebuild" — both options require trusting a third party in addition to Anthropic. A proxy that holds your API keys and intercepts every request isn't obviously a smaller trust surface than a rebuild whose source you can read end-to-end. ClawCodex is fully open source, so the implementation is auditable; the signed binary isn't.

Updates from upstream. This is framed as the rebuild always lagging, but it cuts both ways. When Anthropic ships a breaking change to the binary, a proxy has to scramble to keep up — the rebuild is insulated from that. And the rebuild can ship multi-backend features the official binary will likely never prioritize, since Anthropic has no incentive to make non-Anthropic providers first-class.

Translation fidelity. A proxy re-maps Anthropic's API onto DeepSeek / GLM / MiniMax / OpenAI. That means translating tool use, streaming, system-prompt handling, and caching semantics for each backend. A native multi-provider client calls each backend in its own idiomatic API and avoids that translation layer entirely. This was actually the main reason I switched — fewer edge cases on non-Anthropic models.

Setup. "Wrapper + proxy on :3200" vs a single CLI with a JSON config isn't really a wash — for users who don't want a long-running localhost service in the loop, the rebuild is meaningfully simpler.

Not knocking deepclaude — if keeping the official signed binary is a hard requirement, it's the right call. But "rebuild = inherently riskier" isn't quite right when the rebuild is open source and the proxy is also third-party code in your trust chain.