Deep Research — LangGraph

Turn a fuzzy question into a sourced research brief with open questions and next checks — not a final verdict. (LangGraph stateful graph with a human checkpoint.)

Id: langgraph/deep-research · Slug: deep-research · Status: Untested · Untested on this machine — recipe reviewed for safety (no secrets, no destructive commands, no auto-spend). Mark tested after you run it locally.

What is this

Code for LangGraph. Turn a fuzzy question into a sourced research brief with open questions and next checks — not a final verdict. (LangGraph stateful graph with a human checkpoint.)

How this runtime fits

This LangGraph implementation is a graph-style runtime for the parent AI Team. Agent orchestration and multi-agent workflow language fit when nodes split work and gates; it is not automatically identical to every multi-agent system.

Parent AI Team: deep-research · What is an AI Team?

What it does

Turn a fuzzy question into a sourced research brief with open questions and next checks — not a final verdict. (LangGraph stateful graph with a human checkpoint.)

Who for

Builders working on Research jobs who can run LangGraph themselves.

Need to run

You will run this locally or on infra you control.
No production credentials in prompts or committed files.
Python 3.10+
LangGraph in a venv
Local model binder

How to use

Install LangGraph in a venv (pin versions).
Wire produce() to your local model client.
Run the file once with Example in.
Keep human_gate approval defaulting to False; enable revise only intentionally.

Limitations

Not a substitute for professional legal, medical, or investment advice.
Outputs can be wrong; human review required before publish or spend.
Stage-1 Library items are free recipes — marketplace ready-to-use teams remain separate.
Do not attach billing cloud LLM nodes without a human budget check.

Test status

Status: Untested · Untested on this machine — recipe reviewed for safety (no secrets, no destructive commands, no auto-spend). Mark tested after you run it locally.

Files

graph_deep_research.py — LangGraph nodes + interrupt
README.md — Run notes

Full prompt / config / code

# graph_deep_research.py — Deep Research (LangGraph sketch)
# pip install langgraph langchain-core (pin yourself)
# Use a local chat model binder; do not embed secrets.

from typing import TypedDict, Literal
from langgraph.graph import StateGraph, END

SYSTEM = """You are a careful research desk. Goal: produce a brief, not a verdict.

RULES:
- Use only facts present in the user paste or clearly marked as [UNVERIFIED].
- Separate: Known / Inferred / Unknown / Next checks.
- Cite sources as the user labeled them (URL, doc name, date). Do not invent citations.
- No spending, no shell commands, no credential requests.
- Stop after the brief. Ask one clarifying question only if blocking.

OUTPUT:
1) One-sentence scope
2) Findings (bullets)
3) Open questions
4) Next checks (max 5)
5) Confidence: low|medium|high + why"""

class State(TypedDict):
    user_input: str
    draft: str
    approved: bool
    notes: str

def produce(state: State) -> State:
    # Pseudo: call your local model with SYSTEM + state["user_input"]
    draft = "[MODEL OUTPUT PLACEHOLDER — wire your local LLM here]\n" + state["user_input"][:500]
    return {**state, "draft": draft, "notes": "awaiting human"}

def human_gate(state: State) -> State:
    # In real use: interrupt / input() / UI approval.
    # Default False so nothing auto-publishes.
    return {**state, "approved": False}

def route_after_gate(state: State) -> Literal["done", "revise"]:
    return "done" if state.get("approved") else "done"  # stage-1: always end after gate

g = StateGraph(State)
g.add_node("produce", produce)
g.add_node("human_gate", human_gate)
g.set_entry_point("produce")
g.add_edge("produce", "human_gate")
g.add_conditional_edges("human_gate", route_after_gate, {"done": END, "revise": "produce"})
app = g.compile()

if __name__ == "__main__":
    out = app.invoke({"user_input": "PASTE_EXAMPLE_IN", "draft": "", "approved": False, "notes": ""})
    print(out)

Example in

Question: Why did our signup conversion drop last week?
Sources:
- [A] analytics export 2026-09-01..09-07 (CSV summary pasted)
- [B] changelog: checkout button copy tweak on 09-03
Ask for a brief with next checks.

Example out

Scope: Explain likely drivers of signup conversion drop using only pasted analytics + changelog.
Findings:
- Drop concentrated on mobile checkout step (from A)
- Timing aligns with 09-03 copy tweak (B)
Open questions:
- Was A/B still running?
Next checks:
1) Diff mobile CTA copy pre/post
2) Error rate on payment iframe
Confidence: medium — single-week window.

Model / runtime notes

Graph includes an explicit human checkpoint. Stage-1 ends after gate (no infinite revise loop).

Canonical Team

Parent: deep-research · implementation_id: langgraph/deep-research

GitHub source

teams/deep-research/langgraph · commit eb87e8e049fdf9908e438305ff827c7b617505b5 · license: free-use-at-own-risk

Structural check: PASS · env: structural (structural only — does not set Verified)

Related Library items

Related marketplace Team pages

Soft thematic links only — not identity merges. Marketplace Teams ≠ Library AI Team records.

Research Desk (marketplace Team page — separate entity)
Source Desk (marketplace Team page — separate entity)