文档
供应商

供应商

CC-Relay 通过统一接口支持多个 LLM 供应商。本页介绍如何配置每个供应商。

概述

CC-Relay 作为 Claude Code 和各种 LLM 后端之间的代理。所有供应商都公开 Anthropic 兼容的 Messages API，实现供应商之间的无缝切换。

供应商类型描述成本
Anthropicanthropic直接访问 Anthropic API标准 Anthropic 定价
Z.AIzaiZhipu AI GLM 模型，Anthropic 兼容约为 Anthropic 定价的 1/7
MiniMaxminimaxMiniMax 模型，Anthropic 兼容MiniMax 定价
Ollamaollama本地 LLM 推理免费（本地计算）
AWS Bedrockbedrock通过 AWS 使用 SigV4 认证访问 ClaudeAWS Bedrock 定价
Azure AI Foundryazure通过 Azure MAAS 访问 ClaudeAzure AI 定价
Google Vertex AIvertex通过 Google Cloud 访问 ClaudeVertex AI 定价

Anthropic 供应商

Anthropic 供应商直接连接到 Anthropic 的 API。这是完整访问 Claude 模型的默认供应商。

配置 

providers:
- name: "anthropic"
  type: "anthropic"
  enabled: true
  base_url: "https://api.anthropic.com"  # 可选，使用默认值

  keys:
    - key: "${ANTHROPIC_API_KEY}"
      rpm_limit: 60        # 每分钟请求数
      tpm_limit: 100000    # 每分钟令牌数
      priority: 2          # 更高 = 在故障转移中首先尝试

  models:
    - "claude-sonnet-4-5-20250514"
    - "claude-opus-4-5-20250514"
    - "claude-haiku-3-5-20241022"
[[providers]]
name = "anthropic"
type = "anthropic"
enabled = true
base_url = "https://api.anthropic.com"  # Optional, uses default

[[providers.keys]]
key = "${ANTHROPIC_API_KEY}"
rpm_limit = 60        # Requests per minute
tpm_limit = 100000    # Tokens per minute
priority = 2          # Higher = tried first in failover

models = [
"claude-sonnet-4-5-20250514",
"claude-opus-4-5-20250514",
"claude-haiku-3-5-20241022"
]

API 密钥设置

  1. 在 console.anthropic.com 创建账户
  2. 导航到 Settings > API Keys
  3. 创建新的 API 密钥
  4. 存储在环境变量中: export ANTHROPIC_API_KEY="sk-ant-..."

透明认证支持

Anthropic 供应商支持 Claude Code 订阅用户的透明认证。启用后，cc-relay 会原样转发您的订阅令牌:

server:
auth:
  allow_subscription: true
[server.auth]
allow_subscription = true
# 您的订阅令牌将原样转发
export ANTHROPIC_BASE_URL="http://localhost:8787"
claude

详情请参阅透明认证。

Z.AI 供应商

Z.AI（智谱 AI）通过 Anthropic 兼容 API 提供 GLM 模型。这在保持 API 兼容性的同时提供显著的成本节省（约为 Anthropic 定价的 1/7）。

配置 

providers:
- name: "zai"
  type: "zai"
  enabled: true
  base_url: "https://api.z.ai/api/anthropic"  # 可选，使用默认值

  keys:
    - key: "${ZAI_API_KEY}"
      priority: 1  # 故障转移时优先级低于 Anthropic

  # 将 Claude 模型名称映射到 Z.AI 模型
  model_mapping:
    "claude-sonnet-4-5-20250514": "GLM-4.7"
    "claude-sonnet-4-5": "GLM-4.7"
    "claude-haiku-3-5-20241022": "GLM-4.5-Air"
    "claude-haiku-3-5": "GLM-4.5-Air"

  models:
    - "GLM-4.7"
    - "GLM-4.5-Air"
    - "GLM-4-Plus"
[[providers]]
name = "zai"
type = "zai"
enabled = true
base_url = "https://api.z.ai/api/anthropic"  # Optional, uses default

[[providers.keys]]
key = "${ZAI_API_KEY}"
priority = 1  # Lower priority than Anthropic for failover

# Map Claude model names to Z.AI models
[providers.model_mapping]
"claude-sonnet-4-5-20250514" = "GLM-4.7"
"claude-sonnet-4-5" = "GLM-4.7"
"claude-haiku-3-5-20241022" = "GLM-4.5-Air"
"claude-haiku-3-5" = "GLM-4.5-Air"

models = [
"GLM-4.7",
"GLM-4.5-Air",
"GLM-4-Plus"
]

API 密钥设置

  1. 在 z.ai/model-api 创建账户
  2. 导航到 API Keys 部分
  3. 创建新的 API 密钥
  4. 存储在环境变量中: export ZAI_API_KEY="..."

享受10%折扣: 订阅时使用此邀请链接 — 您和推荐人都可获得10%折扣。

Model Mapping

Model Mapping 将 Anthropic 模型名称转换为 Z.AI 等效模型。当 Claude Code 请求 claude-sonnet-4-5-20250514 时，cc-relay 会自动路由到 GLM-4.7:

model_mapping:
# Claude Sonnet -> GLM-4.7（旗舰模型）
"claude-sonnet-4-5-20250514": "GLM-4.7"
"claude-sonnet-4-5": "GLM-4.7"

# Claude Haiku -> GLM-4.5-Air（快速、经济）
"claude-haiku-3-5-20241022": "GLM-4.5-Air"
"claude-haiku-3-5": "GLM-4.5-Air"
[model_mapping]
# Claude Sonnet -> GLM-4.7 (flagship model)
"claude-sonnet-4-5-20250514" = "GLM-4.7"
"claude-sonnet-4-5" = "GLM-4.7"

# Claude Haiku -> GLM-4.5-Air (fast, economical)
"claude-haiku-3-5-20241022" = "GLM-4.5-Air"
"claude-haiku-3-5" = "GLM-4.5-Air"

成本比较

模型Anthropic（每百万令牌）Z.AI 等效Z.AI 成本
claude-sonnet-4-5$3 输入 / $15 输出GLM-4.7~$0.43 / $2.14
claude-haiku-3-5$0.25 输入 / $1.25 输出GLM-4.5-Air~$0.04 / $0.18

价格为近似值，可能会有变动。

Ollama 供应商

Ollama 通过 Anthropic 兼容 API（Ollama v0.14 以来可用）实现本地 LLM 推理。在本地运行模型以保护隐私、零 API 成本和离线操作。

配置 

providers:
- name: "ollama"
  type: "ollama"
  enabled: true
  base_url: "http://localhost:11434"  # 可选，使用默认值

  keys:
    - key: "ollama"  # Ollama 接受但忽略 API 密钥
      priority: 0    # 故障转移的最低优先级

  # 将 Claude 模型名称映射到本地 Ollama 模型
  model_mapping:
    "claude-sonnet-4-5-20250514": "qwen3:32b"
    "claude-sonnet-4-5": "qwen3:32b"
    "claude-haiku-3-5-20241022": "qwen3:8b"
    "claude-haiku-3-5": "qwen3:8b"

  models:
    - "qwen3:32b"
    - "qwen3:8b"
    - "codestral:latest"
[[providers]]
name = "ollama"
type = "ollama"
enabled = true
base_url = "http://localhost:11434"  # Optional, uses default

[[providers.keys]]
key = "ollama"  # Ollama accepts but ignores API keys
priority = 0    # Lowest priority for failover

# Map Claude model names to local Ollama models
[providers.model_mapping]
"claude-sonnet-4-5-20250514" = "qwen3:32b"
"claude-sonnet-4-5" = "qwen3:32b"
"claude-haiku-3-5-20241022" = "qwen3:8b"
"claude-haiku-3-5" = "qwen3:8b"

models = [
"qwen3:32b",
"qwen3:8b",
"codestral:latest"
]

Ollama 设置

  1. 从 ollama.com 安装 Ollama
  2. 拉取您想使用的模型:
    ollama pull qwen3:32b
ollama pull qwen3:8b
ollama pull codestral:latest
  3. 启动 Ollama（安装时自动运行）

推荐模型

对于 Claude Code 工作流，选择至少 32K 上下文的模型:

模型上下文大小最佳用途
qwen3:32b128K32B 参数通用编码、复杂推理
qwen3:8b128K8B 参数快速迭代、简单任务
codestral:latest32K22B 参数代码生成、专业编码
llama3.2:3b128K3B 参数非常快、基础任务

功能限制

Ollama 的 Anthropic 兼容性是部分的。某些功能不支持:

功能支持备注
Streaming（SSE）是与 Anthropic 相同的事件序列
Tool calling是与 Anthropic 相同的格式
Extended thinking部分budget_tokens 被接受但不强制执行
Prompt caching否cache_control 块被忽略
PDF 输入否不支持
图片 URL否仅支持 Base64 编码
令牌计数否/v1/messages/count_tokens 不可用
tool_choice否无法强制使用特定工具

Docker 网络

在 Docker 中运行 cc-relay 但 Ollama 在主机上时:

providers:
- name: "ollama"
  type: "ollama"
  # 使用 Docker 的主机网关代替 localhost
  base_url: "http://host.docker.internal:11434"
[[providers]]
name = "ollama"
type = "ollama"
# Use Docker's host gateway instead of localhost
base_url = "http://host.docker.internal:11434"

或者使用 --network host 运行 cc-relay:

docker run --network host cc-relay

AWS Bedrock 供应商

AWS Bedrock 通过 Amazon Web Services 提供 Claude 访问，具有企业级安全性和 SigV4 认证。

配置 

providers:
- name: "bedrock"
  type: "bedrock"
  enabled: true

  # AWS region (required)
  aws_region: "us-east-1"

  # Explicit AWS credentials (optional)
  # If not set, uses AWS SDK default credential chain:
  # 1. Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY)
  # 2. Shared credentials file (~/.aws/credentials)
  # 3. IAM role (EC2, ECS, Lambda)
  aws_access_key_id: "${AWS_ACCESS_KEY_ID}"
  aws_secret_access_key: "${AWS_SECRET_ACCESS_KEY}"

  # Map Claude model names to Bedrock model IDs
  model_mapping:
    "claude-sonnet-4-5-20250514": "anthropic.claude-sonnet-4-5-20250514-v1:0"
    "claude-sonnet-4-5": "anthropic.claude-sonnet-4-5-20250514-v1:0"
    "claude-haiku-3-5-20241022": "anthropic.claude-haiku-3-5-20241022-v1:0"

  keys:
    - key: "bedrock-internal"  # Internal key for cc-relay auth
[[providers]]
name = "bedrock"
type = "bedrock"
enabled = true

# AWS region (required)
aws_region = "us-east-1"

# Explicit AWS credentials (optional)
# If not set, uses AWS SDK default credential chain:
# 1. Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY)
# 2. Shared credentials file (~/.aws/credentials)
# 3. IAM role (EC2, ECS, Lambda)
aws_access_key_id = "${AWS_ACCESS_KEY_ID}"
aws_secret_access_key = "${AWS_SECRET_ACCESS_KEY}"

# Map Claude model names to Bedrock model IDs
[providers.model_mapping]
"claude-sonnet-4-5-20250514" = "anthropic.claude-sonnet-4-5-20250514-v1:0"
"claude-sonnet-4-5" = "anthropic.claude-sonnet-4-5-20250514-v1:0"
"claude-haiku-3-5-20241022" = "anthropic.claude-haiku-3-5-20241022-v1:0"

[[providers.keys]]
key = "bedrock-internal"  # Internal key for cc-relay auth

AWS Setup

  1. Enable Bedrock Access: In AWS Console, navigate to Bedrock > Model access and enable Claude models
  2. Configure Credentials: Use one of these methods:
    • Environment Variables: export AWS_ACCESS_KEY_ID=... AWS_SECRET_ACCESS_KEY=...
    • AWS CLI: aws configure
    • IAM Role: Attach Bedrock access policy to EC2/ECS/Lambda role

Bedrock Model IDs

Note: Model IDs change frequently as AWS Bedrock adds new Claude versions. Verify the current list in AWS Bedrock model access documentation before deploying.

Bedrock uses a specific model ID format: anthropic.{model}-v{version}:{minor}

Claude ModelBedrock Model ID
claude-sonnet-4-5-20250514anthropic.claude-sonnet-4-5-20250514-v1:0
claude-opus-4-5-20250514anthropic.claude-opus-4-5-20250514-v1:0
claude-haiku-3-5-20241022anthropic.claude-haiku-3-5-20241022-v1:0

Event Stream Conversion

Bedrock returns responses in AWS Event Stream format. CC-Relay automatically converts this to SSE format for Claude Code compatibility. No additional configuration is needed.

Azure AI Foundry 供应商

Azure AI Foundry 通过 Microsoft Azure 提供 Claude 访问，具有企业级 Azure 集成。

配置 

providers:
- name: "azure"
  type: "azure"
  enabled: true

  # Your Azure resource name (appears in URL: {name}.services.ai.azure.com)
  azure_resource_name: "my-azure-resource"

  # Azure API version (default: 2024-06-01)
  azure_api_version: "2024-06-01"

  # Azure uses x-api-key authentication (Anthropic-compatible)
  keys:
    - key: "${AZURE_API_KEY}"

  # Map Claude model names to Azure deployment names
  model_mapping:
    "claude-sonnet-4-5-20250514": "claude-sonnet-4-5"
    "claude-sonnet-4-5": "claude-sonnet-4-5"
    "claude-haiku-3-5": "claude-haiku-3-5"
[[providers]]
name = "azure"
type = "azure"
enabled = true

# Your Azure resource name (appears in URL: {name}.services.ai.azure.com)
azure_resource_name = "my-azure-resource"

# Azure API version (default: 2024-06-01)
azure_api_version = "2024-06-01"

# Azure uses x-api-key authentication (Anthropic-compatible)
[[providers.keys]]
key = "${AZURE_API_KEY}"

# Map Claude model names to Azure deployment names
[providers.model_mapping]
"claude-sonnet-4-5-20250514" = "claude-sonnet-4-5"
"claude-sonnet-4-5" = "claude-sonnet-4-5"
"claude-haiku-3-5" = "claude-haiku-3-5"

Azure Setup

  1. Create Azure AI Resource: In Azure Portal, create an Azure AI Foundry resource
  2. Deploy Claude Model: Deploy a Claude model in your AI Foundry workspace
  3. Get API Key: Copy the API key from Keys and Endpoint section
  4. Note Resource Name: Your URL is https://{resource_name}.services.ai.azure.com

Deployment Names

Azure uses deployment names as model identifiers. Create deployments in Azure AI Foundry, then map them:

model_mapping:
"claude-sonnet-4-5": "my-sonnet-deployment"  # Your deployment name
[model_mapping]
"claude-sonnet-4-5" = "my-sonnet-deployment"  # Your deployment name

Google Vertex AI 供应商

Vertex AI 通过 Google Cloud 提供 Claude 访问，具有无缝 GCP 集成。

配置 

providers:
- name: "vertex"
  type: "vertex"
  enabled: true

  # Google Cloud project ID (required)
  gcp_project_id: "${GOOGLE_CLOUD_PROJECT}"

  # Google Cloud region (required)
  gcp_region: "us-east5"

  # Map Claude model names to Vertex AI model IDs
  model_mapping:
    "claude-sonnet-4-5-20250514": "claude-sonnet-4-5@20250514"
    "claude-sonnet-4-5": "claude-sonnet-4-5@20250514"
    "claude-haiku-3-5-20241022": "claude-haiku-3-5@20241022"

  keys:
    - key: "vertex-internal"  # Internal key for cc-relay auth
[[providers]]
name = "vertex"
type = "vertex"
enabled = true

# Google Cloud project ID (required)
gcp_project_id = "${GOOGLE_CLOUD_PROJECT}"

# Google Cloud region (required)
gcp_region = "us-east5"

# Map Claude model names to Vertex AI model IDs
[providers.model_mapping]
"claude-sonnet-4-5-20250514" = "claude-sonnet-4-5@20250514"
"claude-sonnet-4-5" = "claude-sonnet-4-5@20250514"
"claude-haiku-3-5-20241022" = "claude-haiku-3-5@20241022"

[[providers.keys]]
key = "vertex-internal"  # Internal key for cc-relay auth

GCP Setup

  1. Enable Vertex AI API: In GCP Console, enable the Vertex AI API
  2. Request Claude Access: Request access to Claude models through Vertex AI Model Garden
  3. Configure Authentication: Use one of these methods:
    • Application Default Credentials: gcloud auth application-default login
    • Service Account: Set GOOGLE_APPLICATION_CREDENTIALS environment variable
    • GCE/GKE: Uses attached service account automatically

Vertex AI Model IDs

Vertex AI uses {model}@{version} format:

Claude ModelVertex AI Model ID
claude-sonnet-4-5-20250514claude-sonnet-4-5@20250514
claude-opus-4-5-20250514claude-opus-4-5@20250514
claude-haiku-3-5-20241022claude-haiku-3-5@20241022

Regions

Available regions for Claude on Vertex AI (check Google Cloud documentation for the complete current list):

  • us-east5 (default)
  • us-central1
  • europe-west1

MiniMax 供应商

MiniMax 通过 Anthropic 兼容的 API 提供大语言模型。MiniMax 以具有竞争力的价格提供适合编码任务的高质量模型。

配置 

providers:
- name: "minimax"
  type: "minimax"
  enabled: true
  base_url: "https://api.minimax.io/anthropic"  # 可选，使用默认值

  keys:
    - key: "${MINIMAX_API_KEY}"
      priority: 1  # 故障转移中优先级低于 Anthropic

  # 将 Claude 模型名称映射到 MiniMax 模型
  model_mapping:
    "claude-opus-4-6": "MiniMax-M2.5"
    "claude-sonnet-4-5-20250514": "MiniMax-M2.5-highspeed"
    "claude-sonnet-4-5": "MiniMax-M2.5-highspeed"
    "claude-haiku-4-5-20251001": "MiniMax-M2.1-highspeed"
    "claude-haiku-4-5": "MiniMax-M2.1-highspeed"

  models:
    - "MiniMax-M2.5"
    - "MiniMax-M2.5-highspeed"
    - "MiniMax-M2.1"
    - "MiniMax-M2.1-highspeed"
    - "MiniMax-M2"
[[providers]]
name = "minimax"
type = "minimax"
enabled = true
base_url = "https://api.minimax.io/anthropic"  # 可选，使用默认值

[[providers.keys]]
key = "${MINIMAX_API_KEY}"
priority = 1  # 故障转移中优先级低于 Anthropic

# 将 Claude 模型名称映射到 MiniMax 模型
[providers.model_mapping]
"claude-opus-4-6" = "MiniMax-M2.5"
"claude-sonnet-4-5-20250514" = "MiniMax-M2.5-highspeed"
"claude-sonnet-4-5" = "MiniMax-M2.5-highspeed"
"claude-haiku-4-5-20251001" = "MiniMax-M2.1-highspeed"
"claude-haiku-4-5" = "MiniMax-M2.1-highspeed"

models = [
"MiniMax-M2.5",
"MiniMax-M2.5-highspeed",
"MiniMax-M2.1",
"MiniMax-M2.1-highspeed",
"MiniMax-M2"
]

API 密钥设置

  1. 在 minimax.io 创建账户
  2. 导航到 API 密钥部分
  3. 创建新的 API 密钥
  4. 保存到环境变量: export MINIMAX_API_KEY="..."

认证

MiniMax 使用 Bearer 令牌认证，而不是 Anthropic 使用的 x-api-key 头。CC-Relay 自动处理此差异 — 无需额外配置。

可用模型

模型描述
MiniMax-M2.5旗舰模型，最高质量
MiniMax-M2.5-highspeedM2.5 的高速变体
MiniMax-M2.1上一代模型
MiniMax-M2.1-highspeedM2.1 的高速变体
MiniMax-M2基础模型

模型映射

模型映射将 Anthropic 模型名称转换为 MiniMax 对应模型:

model_mapping:
# Claude Opus -> MiniMax-M2.5（旗舰）
"claude-opus-4-6": "MiniMax-M2.5"

# Claude Sonnet -> MiniMax-M2.5-highspeed（快速、高质量）
"claude-sonnet-4-5-20250514": "MiniMax-M2.5-highspeed"
"claude-sonnet-4-5": "MiniMax-M2.5-highspeed"

# Claude Haiku -> MiniMax-M2.1-highspeed（快速、经济）
"claude-haiku-4-5-20251001": "MiniMax-M2.1-highspeed"
"claude-haiku-4-5": "MiniMax-M2.1-highspeed"
[model_mapping]
# Claude Opus -> MiniMax-M2.5（旗舰）
"claude-opus-4-6" = "MiniMax-M2.5"

# Claude Sonnet -> MiniMax-M2.5-highspeed（快速、高质量）
"claude-sonnet-4-5-20250514" = "MiniMax-M2.5-highspeed"
"claude-sonnet-4-5" = "MiniMax-M2.5-highspeed"

# Claude Haiku -> MiniMax-M2.1-highspeed（快速、经济）
"claude-haiku-4-5-20251001" = "MiniMax-M2.1-highspeed"
"claude-haiku-4-5" = "MiniMax-M2.1-highspeed"

Cloud Provider Comparison

FeatureBedrockAzureVertex AI
AuthenticationSigV4 (AWS)API KeyOAuth2 (GCP)
Streaming FormatEvent StreamSSESSE
Body TransformYesNoYes
Model in URLYesNoYes
Enterprise SSOAWS IAMEntra IDGCP IAM
RegionsUS, EU, APACGlobalUS, EU

Model Mapping

model_mapping 字段将传入的模型名称转换为供应商特定的模型:

providers:
- name: "zai"
  type: "zai"
  model_mapping:
    # 格式: "传入模型": "供应商模型"
    "claude-sonnet-4-5-20250514": "GLM-4.7"
    "claude-sonnet-4-5": "GLM-4.7"
[[providers]]
name = "zai"
type = "zai"

[providers.model_mapping]
# Format: "incoming-model" = "provider-model"
"claude-sonnet-4-5-20250514" = "GLM-4.7"
"claude-sonnet-4-5" = "GLM-4.7"

当 Claude Code 发送:

{"model": "claude-sonnet-4-5-20250514", ...}

CC-Relay 路由到 Z.AI:

{"model": "GLM-4.7", ...}

映射技巧

  1. 包含版本后缀: 同时映射 claude-sonnet-4-5 和 claude-sonnet-4-5-20250514
  2. 考虑上下文长度: 匹配具有类似能力的模型
  3. 测试质量: 验证输出质量满足您的需求

多供应商设置

为故障转移、成本优化或负载分配配置多个供应商:

providers:
# 主要: Anthropic（最高质量）
- name: "anthropic"
  type: "anthropic"
  enabled: true
  keys:
    - key: "${ANTHROPIC_API_KEY}"
      priority: 2  # 首先尝试

# 次要: Z.AI（成本效益）
- name: "zai"
  type: "zai"
  enabled: true
  keys:
    - key: "${ZAI_API_KEY}"
      priority: 1  # 后备

# 第三: Ollama（本地、免费）
- name: "ollama"
  type: "ollama"
  enabled: true
  keys:
    - key: "ollama"
      priority: 0  # 最后手段

routing:
strategy: failover  # 按优先级顺序尝试供应商
# Primary: Anthropic (highest quality)
[[providers]]
name = "anthropic"
type = "anthropic"
enabled = true

[[providers.keys]]
key = "${ANTHROPIC_API_KEY}"
priority = 2  # Tried first

# Secondary: Z.AI (cost-effective)
[[providers]]
name = "zai"
type = "zai"
enabled = true

[[providers.keys]]
key = "${ZAI_API_KEY}"
priority = 1  # Fallback

# Tertiary: Ollama (local, free)
[[providers]]
name = "ollama"
type = "ollama"
enabled = true

[[providers.keys]]
key = "ollama"
priority = 0  # Last resort

[routing]
strategy = "failover"  # Try providers in priority order

使用此配置:

  1. 请求首先发送到 Anthropic（优先级 2）
  2. 如果 Anthropic 失败（429、5xx），尝试 Z.AI（优先级 1）
  3. 如果 Z.AI 失败，尝试 Ollama（优先级 0）

更多选项请参阅路由策略。

故障排除

连接被拒绝（Ollama）

症状: 连接 Ollama 时 connection refused

原因:

  • Ollama 未运行
  • 端口错误
  • Docker 网络问题

解决方案:

# 检查 Ollama 是否正在运行
ollama list

# 验证端口
curl http://localhost:11434/api/version

# 对于 Docker，使用主机网关
base_url: "http://host.docker.internal:11434"

认证失败（Z.AI）

症状: 从 Z.AI 收到 401 Unauthorized

原因:

  • 无效的 API 密钥
  • 环境变量未设置
  • 密钥未激活

解决方案:

# 验证环境变量已设置
echo $ZAI_API_KEY

# 直接测试密钥
curl -X POST https://api.z.ai/api/anthropic/v1/messages \
  -H "x-api-key: $ZAI_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"GLM-4.7","max_tokens":10,"messages":[{"role":"user","content":"Hi"}]}'

模型未找到

症状: model not found 错误

原因:

  • 模型未在 models 列表中配置
  • 缺少 model_mapping 条目
  • 模型未安装（Ollama）

解决方案:

# 确保模型已列出
models:
- "GLM-4.7"

# 确保映射存在
model_mapping:
"claude-sonnet-4-5": "GLM-4.7"
# Ensure model is listed
models = ["GLM-4.7"]

# Ensure mapping exists
[model_mapping]
"claude-sonnet-4-5" = "GLM-4.7"

对于 Ollama，验证模型已安装:

ollama list
ollama pull qwen3:32b

响应缓慢（Ollama）

症状: Ollama 响应非常慢

原因:

  • 模型对硬件来说太大
  • 未使用 GPU
  • RAM 不足

解决方案:

  • 使用更小的模型（用 qwen3:8b 代替 qwen3:32b）
  • 验证 GPU 已启用: ollama run qwen3:8b --verbose
  • 在推理期间检查内存使用情况

后续步骤

API 参考健康检查与熔断器