Keep your graph. Watch it run.

Sulcus instruments your compiled graph’s normal invoke and stream calls. LangGraph still owns scheduling, nodes, tools, state and checkpoints.

Hosted by Sulcus, LangGraph 1.x

Sulcus adds

A live view of graph execution.

  • Graph, chain and node steps from LangGraph callbacks.
  • Model calls with reported token usage, where the model uses LangChain callbacks.
  • Tool start, completion and failure, where callbacks expose them.
  • An optional per-run token limit that blocks the next model call once reached.
  • Stop control and a bounded event history for each run.

LangGraph keeps

The workflow and its state.

  • Graph definition, routing, scheduling and tool execution.
  • Checkpoint storage and state persistence.
  • Interrupts and what your application does with them.
  • Error handling and side effects in your code.

How it attaches

Around your normal graph calls.

When a project selects LangGraph, Sulcus activates its instrumentation in the run’s container and wraps the invoke, ainvoke, stream and astream calls on compiled graphs. Your application calls its graph as usual.

Observation boundary
Runs the workflow
LangGraph
Instrumented calls
Compiled graph invoke / stream
Token source
Framework- and provider-reported usage
Not collected by instrumentation
Prompts, model outputs, tool payloads

What this integration doesn’t do.

Sulcus doesn’t offer approvals for LangGraph interrupts in hosted runs, and it doesn’t move tool calls into another executor or replace the graph’s own checkpoints. Sulcus is developed against LangGraph 1.x.

The token limit relies on reported usage. A request already in flight can report an overage when it completes. Sulcus then records it and ends the run. What your app prints is kept with the run, so avoid printing sensitive data.

See your first run in Sulcus.