Help Wanted: Community Provider Testing & Contributions / 征集社区提供商测试与贡献

[StringKe]

Help Wanted: Community Provider Testing & Contributions

English

Claudex supports routing Claude Code to multiple AI providers through its translation proxy. However, we don't have API keys for every platform and can't test all providers ourselves.

What We Need

We're looking for community members who can help with:

1. Testing existing providers — Verify that profiles in config.example.toml work correctly with real API keys
2. Adding new providers — Contribute profile configs and translation fixes for untested platforms
3. OAuth subscription testing — Test claudex auth login with different subscription plans (ChatGPT Plus/Pro, Gemini Pro, GitHub Copilot, etc.)
4. Fixing translation edge cases — Some providers have slight API differences that need special handling in src/proxy/translation.rs

Currently Tested (v0.2.0)

Provider	Route	Status	Notes
OpenRouter → Claude	openrouter-claude	✅ Working	Correctly identifies as Claude by Anthropic
OpenRouter → GPT	openrouter-gpt	✅ Working	Response via Claude Code system prompt
OpenRouter → Gemini	openrouter-gemini	✅ Working	Correctly identifies as Gemini by Google
OpenRouter → DeepSeek	openrouter-deepseek	✅ Working	Correctly identifies as DeepSeek-V3
OpenRouter → Grok	openrouter-grok	✅ Working	Correctly identifies as Grok by xAI
OpenRouter → Qwen	openrouter-qwen	✅ Working	Correctly identifies as Qwen3 by Alibaba
OpenRouter → Llama	openrouter-llama	✅ Working	Correctly identifies model as llama-4-maverick
MiniMax (Claude proxy)	minimax	✅ Working	DirectAnthropic passthrough
OpenAI (direct API)	openai	✅ Working	Requires max_tokens = 16384 in profile config
Mistral (direct API)	mistral	✅ Working	mistral-large-latest via api.mistral.ai
Cohere (direct API)	cohere	✅ Working	command-a-03-2025 via compatibility API, requires max_tokens = 8192
Z.AI / GLM (direct API)	glm	✅ Working	GLM-4.6 via api.z.ai
Kimi / Moonshot (direct API)	kimi	✅ Working	Kimi-K2 via api.moonshot.ai
Ollama (local)	ollama	✅ Working	Tested with gpt-oss:20b, localhost:11434
OpenAI (Codex CLI OAuth)	codex-sub	✅ Working	OAuth subscription, auto token refresh
Anthropic (direct API)	anthropic	❌ Key Issue	Organization disabled (not a Claudex bug)

Needs Testing / Contributions

Ranked by developer adoption (based on real-world usage data).

High Priority — Most requested by developers, large user base:

Provider	Type	Base URL	Notes
Groq	OpenAICompatible	https://api.groq.com/openai/v1	Ultra-fast inference, free tier available
Google AI Studio / Gemini	OpenAICompatible	https://generativelanguage.googleapis.com/v1beta/openai	Direct Gemini API (not via OpenRouter)
Azure OpenAI	OpenAICompatible	https://{resource}.openai.azure.com/openai/deployments/{model}/v1	Enterprise deployments
Perplexity AI	OpenAICompatible	https://api.perplexity.ai	Search-augmented generation

Medium Priority — Growing adoption, active developer communities:

Provider	Type	Base URL	Notes
Cerebras	OpenAICompatible	https://api.cerebras.ai/v1	Ultra-fast inference hardware
Together AI	OpenAICompatible	https://api.together.xyz/v1	Open model hosting, fine-tuning
Fireworks AI	OpenAICompatible	https://api.fireworks.ai/inference/v1	Fast open model inference
GitHub Models	OpenAICompatible	https://models.inference.ai.azure.com	Free tier with GitHub account

Low Priority — Niche or region-specific:

Provider	Type	Base URL	Notes
Amazon Bedrock	OpenAICompatible	Requires SDK adapter	Enterprise AWS integration
Cloudflare Workers AI	OpenAICompatible	https://api.cloudflare.com/client/v4/accounts/{id}/ai/v1	Edge inference
Nvidia NIM	OpenAICompatible	https://integrate.api.nvidia.com/v1	GPU-optimized inference
Yi / 零一万物	OpenAICompatible	https://api.lingyiwanwu.com/v1	Yi-Lightning
Baichuan / 百川	OpenAICompatible	https://api.baichuan-ai.com/v1	Chinese LLM
Volcengine / 豆包	OpenAICompatible	https://ark.cn-beijing.volces.com/api/v3	ByteDance Doubao
SiliconFlow	OpenAICompatible	https://api.siliconflow.cn/v1	Chinese model aggregator

Known Issues & Tips

• OpenAI direct API: GPT-4o only supports max_tokens = 16384. Add max_tokens = 16384 to profile config (v0.1.2+).
• Cohere: command-a-03-2025 only supports max_tokens = 8192. Use base_url = "https://api.cohere.ai/compatibility/v1" (the OpenAI-compatible endpoint).
• --no-chrome: Claudex v0.1.2+ automatically injects --no-chrome to avoid Chrome integration conflicts.
• OAuth token refresh: OpenAI (Codex CLI) tokens auto-refresh when expired (v0.1.2+).
• Ollama: Use api_key = "ollama" (any non-empty value works, Ollama doesn't validate keys).

How to Contribute

1. Quick test: Add a profile to your config, run:

   claudex run <profile> -p "hello" --dangerously-skip-permissions --no-session-persistence --no-chrome --disable-slash-commands --tools "" --output-format text

2. Config PR: Add a working profile to config.example.toml
3. Translation fix: Fix API format differences in src/proxy/translation.rs
4. OAuth flow: Test claudex auth login <provider> and report results

Profile Config Template

[[profiles]]
name = "provider-name"
provider_type = "OpenAICompatible"
base_url = "https://api.example.com/v1"
api_key = "your-api-key"
default_model = "model-name"
# max_tokens = 16384  # uncomment if provider has a lower limit
priority = 70
enabled = true

[profiles.models]
haiku = "small-model"
sonnet = "default-model"
opus = "large-model"

[profiles.custom_headers]
[profiles.extra_env]

See CONTRIBUTING.md for development setup.

---

征集社区提供商测试与贡献

中文

Claudex 通过翻译代理支持将 Claude Code 路由到多个 AI 提供商。但是，我们没有所有平台的 API Key，无法独自测试所有提供商。

已测试 (v0.2.0)

提供商	Profile 名	状态	备注
OpenRouter → Claude	openrouter-claude	✅ 正常	正确识别为 Claude / Anthropic
OpenRouter → GPT	openrouter-gpt	✅ 正常	通过 Claude Code 系统提示词响应
OpenRouter → Gemini	openrouter-gemini	✅ 正常	正确识别为 Gemini / Google
OpenRouter → DeepSeek	openrouter-deepseek	✅ 正常	正确识别为 DeepSeek-V3
OpenRouter → Grok	openrouter-grok	✅ 正常	正确识别为 Grok / xAI
OpenRouter → Qwen	openrouter-qwen	✅ 正常	正确识别为 Qwen3 / 阿里云
OpenRouter → Llama	openrouter-llama	✅ 正常	正确识别为 llama-4-maverick
MiniMax (Claude 代理)	minimax	✅ 正常	DirectAnthropic 直通
OpenAI (直连 API)	openai	✅ 正常	需配置 max_tokens = 16384
Mistral (直连 API)	mistral	✅ 正常	mistral-large-latest，api.mistral.ai
Cohere (直连 API)	cohere	✅ 正常	command-a-03-2025，需配置 max_tokens = 8192
Z.AI / GLM (直连 API)	glm	✅ 正常	GLM-4.6，api.z.ai
Kimi / Moonshot (直连 API)	kimi	✅ 正常	Kimi-K2，api.moonshot.ai
Ollama (本地)	ollama	✅ 正常	gpt-oss:20b，localhost:11434
OpenAI (Codex CLI OAuth)	codex-sub	✅ 正常	OAuth 订阅，token 自动刷新
Anthropic (直连 API)	anthropic	❌ Key 问题	组织已禁用（非 Claudex bug）

需要测试（按开发者使用频率排序）

高优先级 — 开发者高频使用：

提供商	类型	Base URL	备注
Groq	OpenAICompatible	https://api.groq.com/openai/v1	超快推理，有免费额度
Google AI Studio / Gemini	OpenAICompatible	https://generativelanguage.googleapis.com/v1beta/openai	Gemini 直连（非 OpenRouter）
Azure OpenAI	OpenAICompatible	https://{resource}.openai.azure.com/...	企业部署
Perplexity AI	OpenAICompatible	https://api.perplexity.ai	搜索增强生成

中优先级 — 增长中的开发者社区：

提供商	类型	Base URL	备注
Cerebras	OpenAICompatible	https://api.cerebras.ai/v1	超快推理硬件
Together AI	OpenAICompatible	https://api.together.xyz/v1	开源模型托管
Fireworks AI	OpenAICompatible	https://api.fireworks.ai/inference/v1	快速推理
GitHub Models	OpenAICompatible	https://models.inference.ai.azure.com	GitHub 账号免费额度

低优先级 — 区域或垂直场景：

提供商	类型	Base URL	备注
Amazon Bedrock	OpenAICompatible	需 SDK 适配	AWS 企业集成
Cloudflare Workers AI	OpenAICompatible	https://api.cloudflare.com/...	边缘推理
Nvidia NIM	OpenAICompatible	https://integrate.api.nvidia.com/v1	GPU 优化推理
Yi / 零一万物	OpenAICompatible	https://api.lingyiwanwu.com/v1	Yi-Lightning
Baichuan / 百川	OpenAICompatible	https://api.baichuan-ai.com/v1	中文大模型
火山引擎 / 豆包	OpenAICompatible	https://ark.cn-beijing.volces.com/api/v3	字节跳动豆包
SiliconFlow	OpenAICompatible	https://api.siliconflow.cn/v1	国内模型聚合

已知问题与提示

• OpenAI 直连: GPT-4o 仅支持 max_tokens = 16384，需在 profile 配置中设置（v0.1.2+）
• Cohere: command-a-03-2025 仅支持 max_tokens = 8192，base_url 使用 https://api.cohere.ai/compatibility/v1（OpenAI 兼容端点）
• --no-chrome: Claudex v0.1.2+ 自动注入 --no-chrome 避免 Chrome 集成冲突
• OAuth token 刷新: OpenAI (Codex CLI) token 过期后自动刷新（v0.1.2+）
• Ollama: api_key 填任意非空值即可（如 "ollama"），Ollama 不校验 key

Profile 配置模板

[[profiles]]
name = "provider-name"
provider_type = "OpenAICompatible"
base_url = "https://api.example.com/v1"
api_key = "your-api-key"
default_model = "model-name"
# max_tokens = 16384  # 如果 provider 有更低的上限则取消注释
priority = 70
enabled = true

[profiles.models]
haiku = "small-model"
sonnet = "default-model"
opus = "large-model"

[profiles.custom_headers]
[profiles.extra_env]

如何贡献

1. 在配置中添加 profile，运行测试命令，反馈结果
2. 提交 config.example.toml 的 PR
3. 在 src/proxy/translation.rs 中修复翻译问题
4. 测试 claudex auth login <provider> 并反馈

详见 CONTRIBUTING.md。