客户端

class qveris.QverisClient(config: QverisConfig | None = None, debug_callback: Callable[[str], None] | None = None, *, credential_provider: CredentialProvider | None = None, http_client: AsyncClient | None = None, transport: AsyncBaseTransport | None = None, limits: Limits | None = None)

Async client for Qveris API.

property rate_limit_retries : int

How many times the client has backed off on a 429/503 so far.

Rate-limit backoff is retried pressure, not failure — surface this rather than counting the retried responses as errors.

核心工作流与失败行为

先从 Discover 返回值选择能力，按当前 schema 构造参数；参数由 Agent 生成、schema 较复杂或成本敏感时先 Probe，再 Call。认证、限流重试耗尽和请求校验错误会抛出 SDK 错误；Call 的业务失败还应检查响应中的 success 与错误字段。search_tools、get_tools_by_ids 和 execute_tool 仅为兼容别名，新代码应使用 discover、inspect、probe 和 call。

python

found = await client.discover("weather data for a city")
tool = found.results[0]
details = await client.inspect(tool.tool_id, search_id=found.search_id)
parameters = {"q": "London"}  # Build from details.results[0].params
quote = await client.probe(tool.tool_id, parameters, checks=["schema", "quote"])
result = await client.call(tool.tool_id, parameters, search_id=found.search_id)
if not result.success:
    raise RuntimeError(result.error_message or "Capability call failed")