LangGraph + Instadash — durable agent, durable data

A LangGraph agent that publishes mid-graph output to an editable Instadash grid, interrupts for human review, and reads back the picks before continuing.

#langgraph#python#agent#hitl

What it builds

✓A two-stage LangGraph agent with interrupt_before for HITL
✓Editable grid with a checkbox action column
✓SQLite checkpointer — state persists across process restarts
✓Agent reads back human edits via /<handle>/<slug>/edits?source=human

The key step

graph = (
  StateGraph(State)
    .add_node("generate", generate_candidates)
    .add_node("publish",  publish_for_review)
    .add_node("read_picks", read_human_picks)
    .add_node("analyse",  deep_analyse)
    .add_edge(START, "generate")
    .add_edge("generate",  "publish")
    .add_edge("publish",   "read_picks")
    .add_edge("read_picks","analyse")
    .add_edge("analyse",   END)
    .compile(checkpointer=sqlite_saver,
             interrupt_before=["read_picks"])
)

note ▸

This is the core of the recipe. The full file (including setup, error handling, and the surrounding scaffolding) lives in the GitHub folder linked below — clone or copy it directly.

Run it

bash⎘ copy

git clone https://github.com/instadashio/instadash-recipes
cd instadash-recipes/langgraph-instadash
pip install -r requirements.txt
cp .env.example .env       # fill in your keys
python agent.py --thread-id demo-1