Watch a research agent work — and see which conclusions the evidence earns.
Point it at an autonomous research agent, or the folder it works in. Every question, experiment, run, result and claim appears as a node the moment it happens, each claim is scored on the evidence beneath it, and Replay shows exactly how the agent got there.
Click a hidden task folder to add it back to the dashboard:
python run.py still running?
Retrying automatically…
The server is up — this page is proof of that. What is missing is a task: a piece of work for the graph to follow. Pick whichever of these matches what you have; each one makes a task appear here within a second or two.
The tasks above came with the install so there is something to open — start with the churn model, the smallest one. Archive them whenever you like. Your own work shows up here as soon as you start a task in one of these ways:
Run this in a second terminal, from the project folder. It streams a worked Credit-Risk investigation through the SDK — about twenty nodes fade in and settle on a green-credibility claim. No key, no network.
python examples/demos/live_demo.py
Restart the server in watch mode. Every file that appears or changes under that folder is classified and turned into a node, so an ordinary working directory becomes an evidence graph with no instrumentation.
python run.py --watch ./my-project
No terminal handy? Use to browse to the folder instead — same thing, without the restart.
Create a task, keep the returned api_key, and
have the agent POST one event per decision. This is sidecar mode: the
graph is written by whatever is doing the work.
curl -X POST http://localhost:8000/api/tasks -H "Content-Type: application/json" -d '{"name":"my task"}'