Outsail LangChain / LangGraph Integration (Python)

Thin adapter package that exposes Outsail's curated Capability Registry tools for LangChain and LangGraph Python applications.

Installation

pip install outsail outsail-langchain

Usage

LangChain Tools

from outsail import OutsailClient
from outsail_langchain import OutsailToolkit
from langgraph.prebuilt import ToolNode

async with OutsailClient(
    api_key="osk_live_...",
) as client:
    toolkit = OutsailToolkit(client)
    tools = toolkit.get_tools(profiles=["research", "read_only"])

    # Use with LangGraph ToolNode
    tool_node = ToolNode(tools)

LangGraph Research Node

from outsail import OutsailClient
from outsail_langchain import create_outsail_research_node
from langgraph.graph import StateGraph

async with OutsailClient(
    api_key="osk_live_...",
) as client:
    builder = StateGraph(MyState)
    builder.add_node("research", create_outsail_research_node(client))
    builder.set_entry_point("research")

Available Tools

All 17 curated agent-safe capabilities are exposed as LangChain tools, organised into three profiles:

Research Profile

• outsail_research_start — Start an asynchronous research job
• outsail_research_status — Poll research job status and results
• outsail_verify — Verify a claim with traceable evidence
• outsail_compare — Compare multiple source sets

Read-only Profile

• outsail_context_pack_get — Retrieve a Context Pack by ID
• outsail_agent_context_get — Retrieve agent-shaped context
• outsail_extract — Extract clean text from a URL
• outsail_map — Create a site map of a website
• outsail_report — Generate a research report in Markdown
• outsail_decision_report — Generate a decision-ready report
• outsail_discover — Discover relevant sources

Monitoring Profile

• outsail_monitor_list — List all source monitors
• outsail_monitor_get — Get monitor details
• outsail_monitor_create — Create a source monitor
• outsail_monitor_run — Trigger a monitor check
• outsail_monitor_pause — Pause a monitor
• outsail_monitor_resume — Resume a monitor

API Reference

OutsailToolkit

class OutsailToolkit:
    def __init__(self, client: OutsailClient): ...

    def get_tools(
        self,
        *,
        profiles: Sequence[Literal["research", "read_only", "monitoring"]] | None = None,
    ) -> list[StructuredTool]: ...

create_outsail_tools

def create_outsail_tools(
    client: OutsailClient,
    *,
    profiles: Sequence[ToolProfile] | None = None,
) -> list[StructuredTool]: ...

create_outsail_research_node

def create_outsail_research_node(
    client: OutsailClient,
    *,
    poll_interval_ms: int = 2000,
    timeout_ms: int = 300_000,
    emit_progress: bool = True,
) -> Callable[[dict[str, Any]], dict[str, Any]]: ...

State Contract (Research Node)

Input state keys:

• research_objective (str, required) — Research objective or question
• research_urls (list[str], optional) — Specific URLs to include
• research_discover (bool, optional) — Auto-discover sources (default: True)
• research_profile (str, optional) — Research profile
• research_depth (int, optional) — 0 = fast, 1 = deep

Output state patch:

• context_pack_id (str | None) — Completed Context Pack ID
• research_status ("completed" | "failed" | "timeout")
• research_error (str | None) — Error message if failed
• research_job_id (str | None) — The job ID

Development

# Install in editable mode with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run tests with coverage
pytest --cov=outsail_langchain

License

MIT