Paste a function, hit run, and step through every call,
line, return, and exception —
with an AI agent explaining what actually happened to the stack and
heap at each one.
Nothing to install — paste code, run it, and step through what the interpreter actually did.
Any function or short script. It runs exactly as written, inside a sandboxed tracer.
sys.settrace captures every call, line, return, and exception as a structured event.
Scrub the tape or use ← → — the agent explains each step as you land on it.
Every explanation is grounded in real interpreter state — not a guess about what a line of code "probably" does.
Every frame and heap object gets rendered as a node in a live graph — you see the call stack grow and shrink, not just printed text.
The explanation agent calls state_diff, ast_scope_lookup, and a concept search tool — it only explains what actually changed.
state_diff checks heap identity, not just value — so "the list changed" and "the name points somewhere new" are never confused.
A small curated knowledge base — mutation vs. rebind, call/return, exception propagation — is embedded with Jina and searched via Qdrant for grounding.
Imports are blocked, builtins are restricted, dunder-escape patterns are rejected — plus a step limit and wall-clock timeout on every run.
Every explanation is validated JSON — title, body, key concept, and UI highlight targets — so the frontend never has to parse free text.
Or start from the default example — either way, you'll see it run.
Open the visualizer