GraphLoupe — a self-hosted LangGraph debugger for VS Code

Debug a LangGraph graph at the node level, right in VS Code. Step through a run, inspect and diff state between nodes, time-travel from any checkpoint, count tokens per LLM node — and even pause a node to paste an answer from any chat (Copilot, ChatGPT, Claude) back in as the model's output. A Python FastAPI sidecar runs your graph in isolation; no account, no license, no cloud — a self-hostable alternative to LangGraph Studio.

Built for: "my agent went down the wrong branch / cost too many tokens / hung on a node — how do I see what happened, step by step?"

📖 Full documentation → https://bygao.github.io/GraphLoupe/ — getting started, and a copy-paste snippet for enabling every feature (checkpointer, manual inference, token economy, run-input form, jump-to-source…).

What it does

• Step debugging — breakpoints on any node, a state snapshot + diff at each step, single-step, and time-travel (fork any checkpoint and re-run). Needs compile(checkpointer=…).
• Manual inference (the differentiator) — a node pauses with interrupt(), you copy the rendered prompt into any chat, paste the answer back, and the graph resumes with zero state loss. Text and tool_call both supported.
• Token economy — per-node prompt/completion counts + a run total (exact when the model reports usage_metadata, else a flagged estimate).
• Visualization — ELK orthogonal auto-layout, nodes coloured by kind (script vs ⚡ llm), conditional branches and loops drawn distinctly, the running node lights up.
• Jump to source — click a node's ↗ to open its function in your editor.
• Health panel — one glance shows what's wired up (checkpointer, input schema, LLM nodes, node source) and what isn't, with the langgraph version and interpreter in use.

The IDE never imports your graph — an isolated sidecar worker does (static AST discovery never runs your code). 100 % local: the only traffic is a localhost WebSocket. No telemetry, no credentials requested or stored.

Install

From the VS Code Marketplace — search "GraphLoupe" in the Extensions view, or code --install-extension byGao.graphloupe.

From a .vsix (sideload) — download a release, or build it (npm install && npm run package), then Extensions view → ⋯ → Install from VSIX….

The sidecar needs Python with the deps in requirements.lock on the interpreter you point GraphLoupe at (the Python extension's selected interpreter, or graphloupe.pythonPath). GraphLoupe creates its own managed environment for the sidecar and runs your graph in your interpreter — see the docs.

Quick start (2 minutes)

1. Open a folder in VS Code and press Ctrl/Cmd+Shift+P → "GraphLoupe: Open Graph Panel".
2. "GraphLoupe: Select Graph" — it AST-scans your project for a graph factory (build_graph / build_app / …) or lets you point at a file + symbol manually. To just try it, set the entry to graphloupe_sidecar.graph:showcase_graph (a built-in graph that exercises every feature).
3. ▶ Run. Nodes light up; at the manual node the Manual tab opens with the prompt — paste an answer to steer the flow. Set a breakpoint (click a node) to pause in State; after a run the Tokens tab shows per-node counts; the Health tab shows what's set up.

See the full getting-started guide for connecting your own graph and enabling each feature.

GraphLoupe vs LangGraph Studio

Both visualize and run a LangGraph. GraphLoupe trades the polish of the hosted tool for node-level depth, a manual-inference escape hatch, and zero-dependency self-hosting.

	GraphLoupe	LangGraph Studio
Where it runs	VS Code extension + local Python sidecar	Desktop/hosted app tied to LangSmith
Account / license	None (MIT, fully local)	LangSmith account
Inspection	Node-internal events (astream_events v2)	Checkpoint-level
Any chat as the model	Yes — paste an answer back to resume a paused node	No
Per-node token economy	Yes, built in	Via LangSmith traces

(Positioning as of 2026, not a live feature audit.)

Trust & safety

100 % local, no telemetry, no phone-home. No API-key field; manual inference uses your own chat session. Discovery never runs your code (static AST); execution happens in an isolated sidecar subprocess. Process isolation guards against buggy/runaway graphs, not deliberately malicious code — point it at graphs you'd run yourself. Full threat model in SECURITY.md.

Contributing / from source

npm install && npm run build, then F5 ("Run GraphLoupe Extension"). See the docs for the architecture and dev workflow. MIT-licensed.