{"repo": "antigmalabs/ante-preview", "files": {"docs-site/docs/antix/concepts/routing.md": {"485f037514ac52cdb36b0a5d467b9d32424832b3": {"t": 1785746548.1169221, "n": 4532, "h": "0a40eb06fcbfed13", "c": "---\ntitle: \"Routing & BYOK\"\ndescription: \"Endpoints, multi-protocol SDK compatibility, and Bring Your Own Key semantics.\"\nsidebar_position: 1\n---\n\n# Routing & BYOK\n\nAntix is a multi-protocol gateway. It accepts requests in OpenAI, Anthropic, and Gemini shapes and normalizes streaming across upstream providers including Anthropic, Google Gemini, Alibaba Qwen, DeepSeek, Zai (GLM), xAI, and OpenAI.\n\n## Supported endpoints\n\n| Endpoint | Method | Protocol |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini native (`:generateContent`, `:streamGenerateContent`) |\n| `/v1beta/models/{action}` | POST | Gemini v1beta native path |\n| `/v1/models`, `/models` | GET | Public model catalog (no auth) |\n| `/v2/model/info` | GET | Catalog with pricing |\n\n`/v1/embeddings`, audio, images, files, fine-tuning, and the batch API are **not** supported.\n\n## Drop-in SDK compatibility\n\nPoint any OpenAI, Anthropic, or Gemini SDK at Antix by changing the base URL and swapping in your Virtual Key. The proxy translates provider-specific quirks and keeps SSE streaming predictable.\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\nA streaming pipeline normalizes SSE events across providers so token deltas, tool calls, and stop reasons arrive in a consistent shape regardless of upstream.\n\n### Claude Code\n\nClaude Code uses the Anthropic SDK internally, so redirecting it at Antix takes one environment variable:\n\n```bash\nexport ANTHROPIC_BASE_URL=\"https://antix.antigma.ai\"\n```\n\nThat's it — Claude Code's SDK reads both and routes all `/v1/messages` traffic to Antix, which passes it through to Anthropic with your platform key substituted.\n\n## Bring Your Own Key (BYOK)\n\nIf you have negotiated direct rates with a provider but still want Antix's observability and routing, use BYOK. Send your provider key in `Authorization` and declare the provider with `X-Antix-Provider`:\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nBYOK traffic is not re-billed by Antix. It is still tracked for observability.\n\n### Supported providers\n\nAntix normalizes requests across these upstream providers. Route to any of them by model name (Antix infers the provider) or explicitly via `X-Antix-Provider`:\n\n| Provider | Notable models | Accepted `X-Antix-Provider` values | Notes |\n|---|---|---|---|\n| **Anthropic** | Claude Sonnet, Opus, Haiku | `anthropic` | Native `/v1/messages` support. |\n| **OpenAI** | GPT-5.x | `openai` | Native `/v1/chat/completions` and `/v1/responses` support. |\n| **Google Gemini** | Gemini 3.x | `google`, `gemini`, `google_ai_studio_gemini` | Native Gemini protocol at `/v1/models/{action}` and `/v1beta`. |\n| **xAI** | Grok | `xai`, `x-ai` | OpenAI-compatible upstream. |\n| **Alibaba Qwen** | `qwen3-max`, `qwen3-coder-plus`, `qwen3-coder-flash`, `qwq-plus`, `qwen-plus`, `qwen-max`, and more | `alibaba`, `qwen`, `dashscope` | Routed via DashScope. No distinctive key prefix — you **must** set `X-Antix-Provider` on BYOK calls. |\n| **DeepSeek** | `deepseek-chat` (V3), `deepseek-reasoner` (R1), `deepseek-v4-pro`, `deepseek-v4-flash` | `deepseek` | OpenAI-compatible upstream API. `deepseek-reasoner` returns a `reasoning_content` field alongside the final answer. |\n| **Zai (GLM)** | `glm-5.2`, `glm-5.1`, `glm-4.7` | `zai` | Zhipu AI's GLM model family, OpenAI-compatible upstream. No distinctive key prefix — you **must** set `X-Antix-Provider` on BYOK calls. |\n\n:::note Provider inference\nWhen the header is omitted, Antix infers the provider from the key prefix (e.g., `sk-ant-…` → Anthropic, `sk-…` → OpenAI). Some keys (like Alibaba/DashScope or Zai) may have no distinctive prefix — you **must** set `X-Antix-Provider` for those requests, otherwise they will fail upstream with `401 Unauthorized`.\n:::\n\nSee [Models API](/antix/concepts/models) to list exactly which models your gateway currently serves.\n"}, "ed5bc5390033dc28b138b80fe7660f50229084e8": {"t": 1785765521.4678807, "n": 4532, "h": "0a40eb06fcbfed13", "c": "---\ntitle: \"Routing & BYOK\"\ndescription: \"Endpoints, multi-protocol SDK compatibility, and Bring Your Own Key semantics.\"\nsidebar_position: 1\n---\n\n# Routing & BYOK\n\nAntix is a multi-protocol gateway. It accepts requests in OpenAI, Anthropic, and Gemini shapes and normalizes streaming across upstream providers including Anthropic, Google Gemini, Alibaba Qwen, DeepSeek, Zai (GLM), xAI, and OpenAI.\n\n## Supported endpoints\n\n| Endpoint | Method | Protocol |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini native (`:generateContent`, `:streamGenerateContent`) |\n| `/v1beta/models/{action}` | POST | Gemini v1beta native path |\n| `/v1/models`, `/models` | GET | Public model catalog (no auth) |\n| `/v2/model/info` | GET | Catalog with pricing |\n\n`/v1/embeddings`, audio, images, files, fine-tuning, and the batch API are **not** supported.\n\n## Drop-in SDK compatibility\n\nPoint any OpenAI, Anthropic, or Gemini SDK at Antix by changing the base URL and swapping in your Virtual Key. The proxy translates provider-specific quirks and keeps SSE streaming predictable.\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\nA streaming pipeline normalizes SSE events across providers so token deltas, tool calls, and stop reasons arrive in a consistent shape regardless of upstream.\n\n### Claude Code\n\nClaude Code uses the Anthropic SDK internally, so redirecting it at Antix takes one environment variable:\n\n```bash\nexport ANTHROPIC_BASE_URL=\"https://antix.antigma.ai\"\n```\n\nThat's it — Claude Code's SDK reads both and routes all `/v1/messages` traffic to Antix, which passes it through to Anthropic with your platform key substituted.\n\n## Bring Your Own Key (BYOK)\n\nIf you have negotiated direct rates with a provider but still want Antix's observability and routing, use BYOK. Send your provider key in `Authorization` and declare the provider with `X-Antix-Provider`:\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nBYOK traffic is not re-billed by Antix. It is still tracked for observability.\n\n### Supported providers\n\nAntix normalizes requests across these upstream providers. Route to any of them by model name (Antix infers the provider) or explicitly via `X-Antix-Provider`:\n\n| Provider | Notable models | Accepted `X-Antix-Provider` values | Notes |\n|---|---|---|---|\n| **Anthropic** | Claude Sonnet, Opus, Haiku | `anthropic` | Native `/v1/messages` support. |\n| **OpenAI** | GPT-5.x | `openai` | Native `/v1/chat/completions` and `/v1/responses` support. |\n| **Google Gemini** | Gemini 3.x | `google`, `gemini`, `google_ai_studio_gemini` | Native Gemini protocol at `/v1/models/{action}` and `/v1beta`. |\n| **xAI** | Grok | `xai`, `x-ai` | OpenAI-compatible upstream. |\n| **Alibaba Qwen** | `qwen3-max`, `qwen3-coder-plus`, `qwen3-coder-flash`, `qwq-plus`, `qwen-plus`, `qwen-max`, and more | `alibaba`, `qwen`, `dashscope` | Routed via DashScope. No distinctive key prefix — you **must** set `X-Antix-Provider` on BYOK calls. |\n| **DeepSeek** | `deepseek-chat` (V3), `deepseek-reasoner` (R1), `deepseek-v4-pro`, `deepseek-v4-flash` | `deepseek` | OpenAI-compatible upstream API. `deepseek-reasoner` returns a `reasoning_content` field alongside the final answer. |\n| **Zai (GLM)** | `glm-5.2`, `glm-5.1`, `glm-4.7` | `zai` | Zhipu AI's GLM model family, OpenAI-compatible upstream. No distinctive key prefix — you **must** set `X-Antix-Provider` on BYOK calls. |\n\n:::note Provider inference\nWhen the header is omitted, Antix infers the provider from the key prefix (e.g., `sk-ant-…` → Anthropic, `sk-…` → OpenAI). Some keys (like Alibaba/DashScope or Zai) may have no distinctive prefix — you **must** set `X-Antix-Provider` for those requests, otherwise they will fail upstream with `401 Unauthorized`.\n:::\n\nSee [Models API](/antix/concepts/models) to list exactly which models your gateway currently serves.\n"}, "e153ed13c7b6ff40d5978ee3801a7c72d0ac4ce1": {"t": 1785765522.7181826, "n": 3287, "h": "bde36f118901ee73", "c": "---\ntitle: \"Routing & BYOK\"\ndescription: \"Endpoints, multi-protocol SDK compatibility, and Bring Your Own Key semantics.\"\nsidebar_position: 1\n---\n\n# Routing & BYOK\n\nAntix is a multi-protocol gateway. It accepts requests in OpenAI, Anthropic, and Gemini shapes and normalizes streaming across upstream providers including Anthropic, Google Gemini, Alibaba Qwen, xAI, and OpenAI.\n\n## Supported endpoints\n\n| Endpoint | Method | Protocol |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini native (`:generateContent`, `:streamGenerateContent`) |\n| `/v1beta/models/{action}` | POST | Gemini v1beta native path |\n| `/v1/models`, `/models` | GET | Public model catalog (no auth) |\n| `/v2/model/info` | GET | Catalog with pricing |\n\n`/v1/embeddings`, audio, images, files, fine-tuning, and the batch API are **not** supported.\n\n## Drop-in SDK compatibility\n\nPoint any OpenAI, Anthropic, or Gemini SDK at Antix by changing the base URL and swapping in your Virtual Key. The proxy translates provider-specific quirks and keeps SSE streaming predictable.\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\nA streaming pipeline normalizes SSE events across providers so token deltas, tool calls, and stop reasons arrive in a consistent shape regardless of upstream.\n\n### Claude Code\n\nClaude Code uses the Anthropic SDK internally, so redirecting it at Antix takes one environment variable:\n\n```bash\nexport ANTHROPIC_BASE_URL=\"https://antix.antigma.ai\"\n```\n\nThat's it — Claude Code's SDK reads both and routes all `/v1/messages` traffic to Antix, which passes it through to Anthropic with your platform key substituted.\n\n## Bring Your Own Key (BYOK)\n\nIf you have negotiated direct rates with a provider but still want Antix's observability and routing, use BYOK. Send your provider key in `Authorization` and declare the provider with `X-Antix-Provider`:\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nBYOK traffic is not re-billed by Antix. It is still tracked for observability.\n\n### Accepted `X-Antix-Provider` values\n\n| Provider | Accepted values |\n|---|---|\n| OpenAI | `openai` |\n| Anthropic | `anthropic` |\n| Google Gemini | `google`, `gemini`, `google_ai_studio_gemini` |\n| xAI | `xai`, `x-ai` |\n| Alibaba / DashScope | `alibaba`, `qwen`, `dashscope` |\n\n:::note Provider inference\nWhen the header is omitted, Antix infers the provider from the key prefix (e.g., `sk-ant-…` → Anthropic, `sk-…` → OpenAI). Alibaba/DashScope keys have no distinctive prefix — you **must** set `X-Antix-Provider` for those requests, otherwise they will fail upstream with `401 Unauthorized`.\n:::\n"}, "f4b8d759b6080f1994ccf48f82c6c1f2e7219365": {"t": 1785765524.0007632, "n": 2934, "h": "e822848b112c603a", "c": "---\ntitle: \"Routing & BYOK\"\ndescription: \"Endpoints, multi-protocol SDK compatibility, and Bring Your Own Key semantics.\"\nsidebar_position: 1\n---\n\n# Routing & BYOK\n\nAntix is a multi-protocol gateway. It accepts requests in OpenAI, Anthropic, and Gemini shapes and normalizes streaming across upstream providers including Anthropic, Google Gemini, Alibaba Qwen, xAI, and OpenAI.\n\n## Supported endpoints\n\n| Endpoint | Method | Protocol |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini native (`:generateContent`, `:streamGenerateContent`) |\n| `/v1beta/models/{action}` | POST | Gemini v1beta native path |\n| `/v1/models`, `/models` | GET | Public model catalog (no auth) |\n| `/v2/model/info` | GET | Catalog with pricing |\n\n`/v1/embeddings`, audio, images, files, fine-tuning, and the batch API are **not** supported.\n\n## Drop-in SDK compatibility\n\nPoint any OpenAI, Anthropic, or Gemini SDK at Antix by changing the base URL and swapping in your Virtual Key. The proxy translates provider-specific quirks and keeps SSE streaming predictable.\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\nA streaming pipeline normalizes SSE events across providers so token deltas, tool calls, and stop reasons arrive in a consistent shape regardless of upstream.\n\n## Bring Your Own Key (BYOK)\n\nIf you have negotiated direct rates with a provider but still want Antix's observability and routing, use BYOK. Send your provider key in `Authorization` and declare the provider with `X-Antix-Provider`:\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nBYOK traffic is not re-billed by Antix. It is still tracked for observability.\n\n### Accepted `X-Antix-Provider` values\n\n| Provider | Accepted values |\n|---|---|\n| OpenAI | `openai` |\n| Anthropic | `anthropic` |\n| Google Gemini | `google`, `gemini`, `google_ai_studio_gemini` |\n| xAI | `xai`, `x-ai` |\n| Alibaba / DashScope | `alibaba`, `qwen`, `dashscope` |\n\n:::note Provider inference\nWhen the header is omitted, Antix infers the provider from the key prefix (e.g., `sk-ant-…` → Anthropic, `sk-…` → OpenAI). Alibaba/DashScope keys have no distinctive prefix — you **must** set `X-Antix-Provider` for those requests, otherwise they will fail upstream with `401 Unauthorized`.\n:::\n"}, "5b2e6eb53f6410981f51472557a2c94f1f91fd7e": {"t": 1785765525.2715085, "n": 2939, "h": "ef8784f4717b3670", "c": "---\ntitle: \"Routing & BYOK\"\ndescription: \"Endpoints, multi-protocol SDK compatibility, and Bring Your Own Key semantics.\"\nsidebar_position: 1\n---\n\n# Routing & BYOK\n\nAntix is a multi-protocol gateway. It accepts requests in OpenAI, Anthropic, and Gemini shapes and normalizes streaming across upstream providers including Anthropic, Google Gemini, Alibaba Qwen, xAI, and OpenAI.\n\n## Supported endpoints\n\n| Endpoint | Method | Protocol |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini native (`:generateContent`, `:streamGenerateContent`) |\n| `/v1beta/models/{action}` | POST | Gemini v1beta native path |\n| `/v1/models`, `/models` | GET | Public model catalog (no auth) |\n| `/v2/model/info` | GET | Catalog with pricing |\n\n`/v1/embeddings`, audio, images, files, fine-tuning, and the batch API are **not** supported.\n\n## Drop-in SDK compatibility\n\nPoint any OpenAI, Anthropic, or Gemini SDK at Antix by changing the base URL and swapping in your Virtual Key. The proxy translates provider-specific quirks and keeps SSE streaming predictable.\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\nA Rust streaming pipeline normalizes SSE events across providers so token deltas, tool calls, and stop reasons arrive in a consistent shape regardless of upstream.\n\n## Bring Your Own Key (BYOK)\n\nIf you have negotiated direct rates with a provider but still want Antix's observability and routing, use BYOK. Send your provider key in `Authorization` and declare the provider with `X-Antix-Provider`:\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nBYOK traffic is not re-billed by Antix. It is still tracked for observability.\n\n### Accepted `X-Antix-Provider` values\n\n| Provider | Accepted values |\n|---|---|\n| OpenAI | `openai` |\n| Anthropic | `anthropic` |\n| Google Gemini | `google`, `gemini`, `google_ai_studio_gemini` |\n| xAI | `xai`, `x-ai` |\n| Alibaba / DashScope | `alibaba`, `qwen`, `dashscope` |\n\n:::note Provider inference\nWhen the header is omitted, Antix infers the provider from the key prefix (e.g., `sk-ant-…` → Anthropic, `sk-…` → OpenAI). Alibaba/DashScope keys have no distinctive prefix — you **must** set `X-Antix-Provider` for those requests, otherwise they will fail upstream with `401 Unauthorized`.\n:::\n"}}, "docs-site/i18n/zh-Hans/docusaurus-plugin-content-docs/current/antix/concepts/routing.md": {"485f037514ac52cdb36b0a5d467b9d32424832b3": {"t": 1785746549.66998, "n": 3492, "h": "096d9efa5499347d", "c": "---\ntitle: \"路由与 BYOK\"\ndescription: \"端点、多协议 SDK 兼容性以及自带密钥（Bring Your Own Key）语义。\"\nsidebar_position: 1\n---\n\n# 路由与 BYOK {#routing--byok}\n\nAntix 是一个多协议网关。它接受 OpenAI、Anthropic 和 Gemini 数据结构的请求，并将跨越上游提供商的流式传输规范化，包括 Anthropic、Google Gemini、Alibaba Qwen、DeepSeek、Zai (GLM)、xAI 和 OpenAI。\n\n## 支持的端点 {#supported-endpoints}\n\n| 端点 | 方法 | 协议 |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini 原生（`:generateContent`, `:streamGenerateContent`） |\n| `/v1beta/models/{action}` | POST | Gemini v1beta 原生路径 |\n| `/v1/models`, `/models` | GET | 公开模型目录（无身份验证） |\n| `/v2/model/info` | GET | 包含定价的目录 |\n\n**不**支持 `/v1/embeddings`、音频、图像、文件、微调（fine-tuning）和批处理（batch）API。\n\n## 即插即用的 SDK 兼容性 {#drop-in-sdk-compatibility}\n\n通过更改基地址（base URL）并换上您的虚拟密钥，将任何 OpenAI、Anthropic 或 Gemini SDK 指向 Antix。代理会转换特定于提供商的特性，并保持 SSE 流的传输可预测。\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\n流式传输管道将跨提供商的 SSE 事件标准化，因此无论上游情况如何，代币增量、工具调用和停止原因都将以一致的数据结构到达。\n\n### Claude Code {#claude-code}\n\nClaude Code 内部使用 Anthropic SDK，因此将其重定向到 Antix 只需要一个环境变量：\n\n```bash\nexport ANTHROPIC_BASE_URL=\"https://antix.antigma.ai\"\n```\n\n就是这样——Claude Code 的 SDK 会读取这两个配置，并将所有的 `/v1/messages` 流量路由到 Antix，Antix 在替换您的平台密钥后将其原样传递给 Anthropic。\n\n## 自带密钥（Bring Your Own Key, BYOK） {#bring-your-own-key-byok}\n\n如果您已经与提供商协商了直接费率，但仍然想要 Antix 的可观测性和路由功能，请使用 BYOK。在 `Authorization` 中发送您的提供商密钥，并通过 `X-Antix-Provider` 声明提供商：\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nAntix 不会对 BYOK 流量重新计费。为了可观测性，它仍会被追踪记录。\n\n### 支持的提供商 {#supported-providers}\n\nAntix 在这些上游提供商之间将请求标准化。您可以通过模型名称（Antix 会推断提供商）或通过 `X-Antix-Provider` 明确地路由到其中的任何一个：\n\n| 提供商 | 显著的模型 | 可接受的 `X-Antix-Provider` 值 | 备注 |\n|---|---|---|---|\n| **Anthropic** | Claude Sonnet, Opus, Haiku | `anthropic` | 原生 `/v1/messages` 支持。 |\n| **OpenAI** | GPT-5.x | `openai` | 原生 `/v1/chat/completions` 和 `/v1/responses` 支持。 |\n| **Google Gemini** | Gemini 3.x | `google`, `gemini`, `google_ai_studio_gemini` | `/v1/models/{action}` 和 `/v1beta` 的原生 Gemini 协议。 |\n| **xAI** | Grok | `xai`, `x-ai` | 兼容 OpenAI 的上游。 |\n| **Alibaba Qwen** | `qwen3-max`, `qwen3-coder-plus`, `qwen3-coder-flash`, `qwq-plus`, `qwen-plus`, `qwen-max`, 等等 | `alibaba`, `qwen`, `dashscope` | 通过 DashScope 路由。没有独特的密钥前缀——您**必须**在 BYOK 调用中设置 `X-Antix-Provider`。 |\n| **DeepSeek** | `deepseek-chat` (V3), `deepseek-reasoner` (R1), `deepseek-v4-pro`, `deepseek-v4-flash` | `deepseek` | 兼容 OpenAI 的上游 API。`deepseek-reasoner` 会在最终答案旁返回 `reasoning_content` 字段。 |\n| **Zai (GLM)** | `glm-5.2`, `glm-5.1`, `glm-4.7` | `zai` | 智谱 AI 的 GLM 模型系列，兼容 OpenAI 的上游。没有独特的密钥前缀——您**必须**在 BYOK 调用中设置 `X-Antix-Provider`。 |\n\n:::note 提供商推断\n当标头被省略时，Antix 会从密钥前缀推断提供商（例如，`sk-ant-…` → Anthropic，`sk-…` → OpenAI）。某些密钥（如 Alibaba/DashScope 或 Zai）可能没有明显的前缀——您**必须**为这些请求设置 `X-Antix-Provider`，否则它们在上游将因 `401 Unauthorized` 失败。\n:::\n\n请参阅[模型 API](/antix/concepts/models)以准确列出您的网关目前提供服务的模型。\n"}, "f2861ee4088de1e56a31265242f7d190cb336a35": {"t": 1785765521.1335819, "n": 3492, "h": "096d9efa5499347d", "c": "---\ntitle: \"路由与 BYOK\"\ndescription: \"端点、多协议 SDK 兼容性以及自带密钥（Bring Your Own Key）语义。\"\nsidebar_position: 1\n---\n\n# 路由与 BYOK {#routing--byok}\n\nAntix 是一个多协议网关。它接受 OpenAI、Anthropic 和 Gemini 数据结构的请求，并将跨越上游提供商的流式传输规范化，包括 Anthropic、Google Gemini、Alibaba Qwen、DeepSeek、Zai (GLM)、xAI 和 OpenAI。\n\n## 支持的端点 {#supported-endpoints}\n\n| 端点 | 方法 | 协议 |\n|---|---|---|\n| `/v1/chat/completions` | POST | OpenAI Chat Completions |\n| `/v1/responses` | POST | OpenAI Responses API |\n| `/v1/messages` | POST | Anthropic Messages |\n| `/v1/messages/count_tokens` | POST | Anthropic token counter |\n| `/v1/models/{action}` | POST | Gemini 原生（`:generateContent`, `:streamGenerateContent`） |\n| `/v1beta/models/{action}` | POST | Gemini v1beta 原生路径 |\n| `/v1/models`, `/models` | GET | 公开模型目录（无身份验证） |\n| `/v2/model/info` | GET | 包含定价的目录 |\n\n**不**支持 `/v1/embeddings`、音频、图像、文件、微调（fine-tuning）和批处理（batch）API。\n\n## 即插即用的 SDK 兼容性 {#drop-in-sdk-compatibility}\n\n通过更改基地址（base URL）并换上您的虚拟密钥，将任何 OpenAI、Anthropic 或 Gemini SDK 指向 Antix。代理会转换特定于提供商的特性，并保持 SSE 流的传输可预测。\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI(\n    base_url=\"https://antix.antigma.ai/v1\",\n    api_key=\"sk-antix-<your-key>\",\n)\n\nresponse = client.chat.completions.create(\n    model=\"claude-sonnet-4-6\",\n    messages=[{\"role\": \"user\", \"content\": \"Hello!\"}],\n    stream=True,\n)\n```\n\n流式传输管道将跨提供商的 SSE 事件标准化，因此无论上游情况如何，代币增量、工具调用和停止原因都将以一致的数据结构到达。\n\n### Claude Code {#claude-code}\n\nClaude Code 内部使用 Anthropic SDK，因此将其重定向到 Antix 只需要一个环境变量：\n\n```bash\nexport ANTHROPIC_BASE_URL=\"https://antix.antigma.ai\"\n```\n\n就是这样——Claude Code 的 SDK 会读取这两个配置，并将所有的 `/v1/messages` 流量路由到 Antix，Antix 在替换您的平台密钥后将其原样传递给 Anthropic。\n\n## 自带密钥（Bring Your Own Key, BYOK） {#bring-your-own-key-byok}\n\n如果您已经与提供商协商了直接费率，但仍然想要 Antix 的可观测性和路由功能，请使用 BYOK。在 `Authorization` 中发送您的提供商密钥，并通过 `X-Antix-Provider` 声明提供商：\n\n```bash\ncurl -X POST https://antix.antigma.ai/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -H \"Authorization: Bearer YOUR_ALIBABA_DASHSCOPE_KEY\" \\\n  -H \"X-Antix-Provider: alibaba\" \\\n  -d '{\n    \"model\": \"qwen-max\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}]\n  }'\n```\n\nAntix 不会对 BYOK 流量重新计费。为了可观测性，它仍会被追踪记录。\n\n### 支持的提供商 {#supported-providers}\n\nAntix 在这些上游提供商之间将请求标准化。您可以通过模型名称（Antix 会推断提供商）或通过 `X-Antix-Provider` 明确地路由到其中的任何一个：\n\n| 提供商 | 显著的模型 | 可接受的 `X-Antix-Provider` 值 | 备注 |\n|---|---|---|---|\n| **Anthropic** | Claude Sonnet, Opus, Haiku | `anthropic` | 原生 `/v1/messages` 支持。 |\n| **OpenAI** | GPT-5.x | `openai` | 原生 `/v1/chat/completions` 和 `/v1/responses` 支持。 |\n| **Google Gemini** | Gemini 3.x | `google`, `gemini`, `google_ai_studio_gemini` | `/v1/models/{action}` 和 `/v1beta` 的原生 Gemini 协议。 |\n| **xAI** | Grok | `xai`, `x-ai` | 兼容 OpenAI 的上游。 |\n| **Alibaba Qwen** | `qwen3-max`, `qwen3-coder-plus`, `qwen3-coder-flash`, `qwq-plus`, `qwen-plus`, `qwen-max`, 等等 | `alibaba`, `qwen`, `dashscope` | 通过 DashScope 路由。没有独特的密钥前缀——您**必须**在 BYOK 调用中设置 `X-Antix-Provider`。 |\n| **DeepSeek** | `deepseek-chat` (V3), `deepseek-reasoner` (R1), `deepseek-v4-pro`, `deepseek-v4-flash` | `deepseek` | 兼容 OpenAI 的上游 API。`deepseek-reasoner` 会在最终答案旁返回 `reasoning_content` 字段。 |\n| **Zai (GLM)** | `glm-5.2`, `glm-5.1`, `glm-4.7` | `zai` | 智谱 AI 的 GLM 模型系列，兼容 OpenAI 的上游。没有独特的密钥前缀——您**必须**在 BYOK 调用中设置 `X-Antix-Provider`。 |\n\n:::note 提供商推断\n当标头被省略时，Antix 会从密钥前缀推断提供商（例如，`sk-ant-…` → Anthropic，`sk-…` → OpenAI）。某些密钥（如 Alibaba/DashScope 或 Zai）可能没有明显的前缀——您**必须**为这些请求设置 `X-Antix-Provider`，否则它们在上游将因 `401 Unauthorized` 失败。\n:::\n\n请参阅[模型 API](/antix/concepts/models)以准确列出您的网关目前提供服务的模型。\n"}}}}