Data source

Every screen reads from one event stream. Replay plays recorded events, so UI work and the demo fallback never wait on the backend. Live reads the coordinator.

Live connection

Events from the coordinator's job stream

Not connected

Stream: https://gpuchainv1.duckdns.org/sepolia
REST: https://gpuchainv1.duckdns.org/sepolia
Last message: none yet

Not connected? Start the stack

From the repo root, start the CPU stack (no GPU, no Docker):
```
# Windows
.cache\cpu-venv\Scripts\python.exe experiments\demo_stack.py --cpu --no-docker --total-steps 60000
# macOS / Linux
.cache/cpu-venv/bin/python experiments/demo_stack.py --cpu --no-docker --total-steps 60000
```
The GPU stack (experiments/demo_stack.py without --cpu) uses the real GPU, and --no-docker runs it without Docker. 60000 steps leaves time to press Kill; 6000 finishes in seconds on a fast CPU.
Wait for it to print the API address. It should be https://gpuchainv1.duckdns.org/sepolia.
This page reconnects by itself within a few seconds. Or press Retry now.

Right now

The plain-language line mission control leads with

No job yet

Waiting for the first event

Job state

unknown

Fencing token

unknown

Latest step

unknown

Event stream

Newest first, last 40. Open a row for its raw payload

No events yet

Waiting for the first event from the coordinator.

Loading page…