Function paper_flow

quantmind/flows/paper.py:64–114

Extract a ``Paper`` from a typed ``PaperInput``. See design doc §4.1 for the rationale on each kwarg. ``memory`` is a PR6 placeholder — non-None values are accepted but unused in PR5. Raises: UnsupportedContentTypeError: When fetched bytes are not PDF / HTML / markd

(
    input: PaperInput,
    *,
    cfg: PaperFlowCfg | None = None,
    extra_tools: list[Tool] | None = None,
    extra_instructions: str | None = None,
    output_type: type[P] | None = None,
    memory: object | None = None,
    extra_run_hooks: list[RunHooks[Any]] | None = None,
    extra_input_guardrails: list[Any] | None = None,
    extra_output_guardrails: list[Any] | None = None,
)

async def paper_flow(
    input: PaperInput,
    *,
    cfg: PaperFlowCfg | None = None,
    extra_tools: list[Tool] | None = None,
    extra_instructions: str | None = None,
    output_type: type[P] | None = None,
    memory: object | None = None,
    extra_run_hooks: list[RunHooks[Any]] | None = None,
    extra_input_guardrails: list[Any] | None = None,
    extra_output_guardrails: list[Any] | None = None,
) -> P | Paper:
    """Extract a ``Paper`` from a typed ``PaperInput``.

    See design doc §4.1 for the rationale on each kwarg. ``memory`` is a
    PR6 placeholder — non-None values are accepted but unused in PR5.

    Raises:
        UnsupportedContentTypeError: When fetched bytes are not PDF /
            HTML / markdown / plain-text.
        NotImplementedError: When ``input`` is a ``DoiIdentifier`` (the
            unpaywall fallback is its own follow-up issue).
    """
    cfg = cfg or PaperFlowCfg()
    out_type: type[Paper] = output_type or Paper  # type: ignore[assignment]

    raw_md, source_meta = await _fetch_and_format(input)

    # Agent's `model_settings` parameter is non-optional (defaults to a
    # fresh ``ModelSettings()``); only forward when cfg has one set.
    agent_kwargs: dict[str, Any] = {
        "name": "paper_extractor",
        "instructions": _compose_instructions(
            _DEFAULT_INSTRUCTIONS, extra_instructions, cfg
        ),
        "model": cfg.model,
        "tools": list(extra_tools or []),
        "output_type": out_type,
        "input_guardrails": list(extra_input_guardrails or []),
        "output_guardrails": list(extra_output_guardrails or []),
    }
    if cfg.model_settings is not None:
        agent_kwargs["model_settings"] = cfg.model_settings
    agent: Agent[Any] = Agent(**agent_kwargs)
    return await run_with_observability(
        agent,
        _format_input(raw_md, source_meta),
        cfg=cfg,
        memory=memory,
        extra_run_hooks=list(extra_run_hooks or []),
    )


async def _fetch_and_format(