Basic Info

客户端调用时通过 x-portkey-config: claude-fb-prod

Routing Strategy

选择请求分发模式
Single
固定 1 个目标
Fallback
失败回退到下一个
Loadbalance
按权重分流
Conditional
按条件路由

Targets

Fallback 顺序:1 → 2 → 3
1
Primary
2
Fallback on: 429, 503, timeout
3
Last resort

Reliability

Retry on failure
同一 target 重试次数(429/5xx 触发)
Request timeout (ms)
单个 target 超时阈值
Circuit Breaker
连续失败 N 次自动跳过该 target
Semantic Cache
相似 prompt 命中缓存(Phase 2 上线)

Guardrails (Optional)

提示:可在请求 / 响应阶段插入 Guardrail 检查(如 PII 检测、内容过滤)。
Generated config.json
⧉ Copy
{
  "name": "claude-fallback-prod",
  "strategy": {
    "mode": "fallback",
    "on_status_codes": [429, 503],
    "on_timeout": true
  },
  "targets": [
    {
      "virtual_key": "acai-anthropic",
      "override_params": {
        "model": "claude-opus-4-7"
      }
    },
    {
      "virtual_key": "acai-anthropic",
      "override_params": {
        "model": "claude-sonnet-4-6"
      }
    },
    {
      "virtual_key": "openai-hk-primary",
      "override_params": {
        "model": "gpt-4o"
      }
    }
  ],
  "retry": {
    "attempts": 3,
    "on_status_codes": [429, 500, 502, 503]
  },
  "request_timeout": 60000,
  // circuit_breaker disabled
  // cache disabled (Phase 2)
  // guardrails: none
  "metadata": {
    "workspace": "wagela-prod",
    "author": "jerry.du",
    "updated_at": "2026-06-04T03:15Z"
  }
}
Estimated behaviour
客户端调用 → 先打 Opus 4.7
429/503/超时 → 落 Sonnet 4.6
再失败 → 落 GPT-4o (HK)
每跳最多重试 3 次,单跳 ≤ 60s