Live execution graph
Nodes light up as they start and stay marked once they're done, so the path through the graph is visible while it happens.
See what your agent actually did.
A live visualization for LangGraph and DeepAgents. Wrap your compiled graph in one call and watch the run unfold in your browser: which branches it took, where it looped, and where it stopped.
Install LangGraphics
$ pip install langgraphics
One line
Wrap the graph. Nothing else changes.
from langgraph.graph import StateGraph, MessagesState
from langgraphics import watch
workflow = StateGraph(MessagesState)
workflow.add_node(...)
workflow.add_edge(...)
graph = watch(workflow.compile())
await graph.ainvoke({"messages": [...]})
What you get
A run you can read, not a log you have to.
One-line setup
No refactoring and no decorators to sprinkle around. One call on the compiled graph, and your agent returns exactly what it did before.
Inspect panel
Open a node to read its input, its output and the slice of state it produced, as a tree or in full.
Replay
A run is recorded as it happens, so you can play it back step by step as many times as you need to work out what went wrong.
Subgraphs in place
Nested graphs are drawn inside their parent, so a DeepAgents run with sub-agents reads as easily as a flat one.
Cost and latency
Per-node latency, token counts and spend, cached and total, next to the node that ran up the bill.
Entirely local
An HTTP server and a WebSocket on your own machine. No account, no API key, and nothing about your run leaves the process.
Failures in context
When a node raises, it's marked where it sits in the graph, with everything that led to it still on screen.
Where it fits
Next to the hosted tracers.
| Capability | LangGraphics | LangFuse | LangSmith |
|---|---|---|---|
| Fully local | |||
| Data stays local | |||
| No API key required | |||
| One-line setup | |||
| Live execution graph | |||
| Subgraph visualization | |||
| Replay mode | |||
| Self-hosted | |||
| Cost and latency tracking | |||
| Prompt evaluation |