Agently 文档

中文

English

中文

English

Appearance

Provider 配置 ​

这一页提供常见 provider 的配置形状。把对应代码块放进 Agently.set_settings(...)，再按 provider 官方文档和团队环境填当前可用模型名。

所有密钥都用 ${ENV.*} 占位，避免把 token 写进代码。协议选择逻辑见 模型概览。

OpenAI ​

Chat Completions：

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://api.openai.com/v1",
    "api_key": "${ENV.OPENAI_API_KEY}",
    "model": "${ENV.OPENAI_MODEL}",
})

Responses API：

python
Agently.set_settings("OpenAIResponsesCompatible", {
    "base_url": "https://api.openai.com/v1",
    "api_key": "${ENV.OPENAI_API_KEY}",
    "model": "${ENV.OPENAI_RESPONSES_MODEL}",
})

Anthropic / Claude ​

python
Agently.set_settings("AnthropicCompatible", {
    "base_url": "https://api.anthropic.com",
    "api_key": "${ENV.ANTHROPIC_API_KEY}",
    "model": "${ENV.ANTHROPIC_MODEL}",
    "max_tokens": 4096,
})

anthropic_version、anthropic_beta 等字段见 AnthropicCompatible。

DeepSeek ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://api.deepseek.com/v1",
    "api_key": "${ENV.DEEPSEEK_API_KEY}",
    "model": "${ENV.DEEPSEEK_MODEL}",
})

Qwen / DashScope ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
    "api_key": "${ENV.DASHSCOPE_API_KEY}",
    "model": "${ENV.DASHSCOPE_MODEL}",
})

Kimi / Moonshot ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://api.moonshot.cn/v1",
    "api_key": "${ENV.MOONSHOT_API_KEY}",
    "model": "${ENV.MOONSHOT_MODEL}",
})

GLM / 智谱 ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://open.bigmodel.cn/api/paas/v4/",
    "api_key": "${ENV.GLM_API_KEY}",
    "model": "${ENV.GLM_MODEL}",
})

MiniMax ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://api.minimax.chat/v1",
    "api_key": "${ENV.MINIMAX_API_KEY}",
    "model": "${ENV.MINIMAX_MODEL}",
})

Doubao（豆包） ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://ark.cn-beijing.volces.com/api/v3",
    "api_key": "${ENV.DOUBAO_API_KEY}",
    "model": "${ENV.DOUBAO_MODEL}",
})

ERNIE（文心） ​

千帆 OpenAI 兼容模式：

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://qianfan.baidubce.com/v2",
    "api_key": "${ENV.QIANFAN_API_KEY}",
    "model": "${ENV.QIANFAN_MODEL}",
})

SiliconFlow（硅基流动） ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://api.siliconflow.cn/v1",
    "api_key": "${ENV.SILICONFLOW_API_KEY}",
    "model": "${ENV.SILICONFLOW_MODEL}",
})

Groq ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://api.groq.com/openai/v1",
    "api_key": "${ENV.GROQ_API_KEY}",
    "model": "${ENV.GROQ_MODEL}",
})

Gemini（经 OpenAI 兼容端点） ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
    "api_key": "${ENV.GEMINI_API_KEY}",
    "model": "${ENV.GEMINI_MODEL}",
})

Ollama（本地） ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "http://127.0.0.1:11434/v1",
    "model": "qwen2.5:7b",
})

本地服务无鉴权时可以省略 api_key。模型名要和本机 ollama pull 后可用的名字一致。

vLLM / LM Studio / llama.cpp server（本地） ​

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "http://localhost:8000/v1",
    "model": "your-served-model-name",
})

端口、模型名和鉴权方式取决于本地服务启动参数。

自建内部网关 ​

如果团队私有网关说 OpenAI Chat Completions：

python
Agently.set_settings("OpenAICompatible", {
    "base_url": "https://gw.internal/openai-compat/v1",
    "api_key": "${ENV.INTERNAL_GATEWAY_TOKEN}",
    "model": "internal-default",
})

如果网关说 Anthropic Messages API，换成 AnthropicCompatible，并补上 max_tokens：

python
Agently.set_settings("AnthropicCompatible", {
    "base_url": "https://gw.internal/anthropic",
    "api_key": "${ENV.INTERNAL_GATEWAY_TOKEN}",
    "model": "internal-claude",
    "max_tokens": 4096,
})

配好以后跑一个 smoke ​

配置完成后，用目标 provider 跑一次最小结构化输出：

python
agent = Agently.create_agent()
result = (
    agent
    .input("返回 OK。")
    .output({"answer": (str, "回答", True)})
    .get_result()
)
data = await result.async_get_data(ensure_keys=["answer"])

这个 smoke 能同时验证 base URL、key、model、网络、结构化输出解析。不要只靠配置文件看起来正确。

另见 ​