Skip to main content

Module airbyte_agent_sdk.tools

Framework-ready connector tools.

Functions​

build_agent_tool_decorator(model: Any, *, role: AgentToolRole | None = None, inspect_tool: str | None = None, docs_tool: str | None = None, max_output_chars: int | None | Unset = UNSET, framework: FrameworkName = 'none', internal_retries: int = 0, should_internal_retry: Callable[[Exception, tuple[Any, ...], dict[str, Any]], bool] | None = None, exhausted_runtime_failure_message: Callable[[Exception, tuple[Any, ...], dict[str, Any]], str | None] | None = None) ‑> Callable[[Callable[..., typing.Any]], Callable[..., typing.Any]] : Build the agent_tool decorator bound to one connector model.

Backs the generated @<Connector>.agent_tool(...) classmethod, which documents the user-facing contract. Always uses progressive skill docs: the execute docstring points the agent at the connector's inspect and docs tools instead of embedding the full entity/action reference.

build_connector_tools(connector: Any, *, framework: FrameworkName | None = None, docs_provider: ConnectorDocsProvider | None = None, use_progressive_docs: bool = True, max_output_chars: int | None = 100000, internal_retries: int = 0, should_internal_retry: Callable[[Exception, tuple[Any, ...], dict[str, Any]], bool] | None = None, exhausted_runtime_failure_message: Callable[[Exception, tuple[Any, ...], dict[str, Any]], str | None] | None = None) ‑> airbyte_agent_sdk.tools.ConnectorTools : Build inspect/docs/execute tools bound to a single connector.

Hosted connectors use the live inspect and skill-docs endpoints. Local connectors keep the generated YAML-derived rich docs as their fallback.

Classes​

ConnectorDocsProvider(*args, **kwargs) : Provider of connector inspection and skill-doc endpoints.

Ancestors (in MRO)​

  • typing.Protocol
  • typing.Generic

Methods​

inspect_connector(self) ‑> dict[str, typing.Any] :

read_skill_docs(self, id: str, section: str | None = None) ‑> dict[str, typing.Any] :

ConnectorTools(inspect_connector: ToolCallable, read_skill_docs: ToolCallable, execute: ToolCallable, use_progressive_docs: bool = True) : Connector tool callables for agent frameworks.

Instance variables​

execute: Callable[..., typing.Awaitable[typing.Any]] : The type of the None singleton.

inspect_connector: Callable[..., typing.Awaitable[typing.Any]] : The type of the None singleton.

read_skill_docs: Callable[..., typing.Awaitable[typing.Any]] : The type of the None singleton.

use_progressive_docs: bool : The type of the None singleton.

Methods​

as_list(self) ‑> list[Callable[..., typing.Awaitable[typing.Any]]] :

SkillDocsAccessor(connector: Any, *, docs_provider: ConnectorDocsProvider | None = None) : Stateful access to one connector's skill docs.

Owns the docs_skill_id cache: populated after the first successful :meth:inspect, re-inspected only when missing, never populated on failure. Cache lifetime is the accessor instance — build_connector_tools shares one across its tool closures; generated typed connectors hold one per connector instance.

Instance variables​

provider: ConnectorDocsProvider | None :

Methods​

inspect(self) ‑> dict[str, typing.Any] : Inspect the connector's hosted metadata and resolve its docs skill id.

Local/offline connectors (no docs provider) get a local-mode payload with a warning instead of a hosted inspection.

read(self, section: str | None = None) ‑> str : Read the connector's usage docs, rendered to text.

Omit section for the outline and general guidance; pass an exact section id for full details. Local/offline connectors return the full generated docs and ignore section.

Unset() : Sentinel type marking 'argument not provided' (distinct from None).

Was this page helpful?