Chat in use
You can continue when they’re done
Public. Shared with everyone.
Join
Contribute this device
Experimental public chat. Messages are shared with everyone in the pool; do not enter sensitive information.
Optional compute · Light by default. Downloads ≈72 MiB for one complete model. Uses memory, power and data. Stop releases resources. No actual payouts.
Devices
—
+0
Memory
+0 MiB
— loaded
Pool · session
$0.00
+$0.00
This device · session
Light takes breaks. Pauses while hidden or busy. One contributing tab per browser profile.
Join to chat
— loading
Starting…
Join ↗
Theoretical cloud value. No actual payout.
Waiting for rates…
Dense FP32 work × cloud cost per FLOP. Hourly estimates use this device’s measured productive app work ÷ successful client execution time, including client overhead, at 100% activity, separate from its memory slider. They do not measure its physical maximum. Month = 40 h × 4 weeks (160 h). No fees, power costs or demand allowance. The GPU peak reference is theoretical, not a workload benchmark. Downloads, idle time and standby earn zero. Finished work counts even if a run later fails.
Device totals cover this connection.
The app maximum uses a conservative policy and any RAM estimate. Browsers do not expose free RAM or GPU memory. Actual allocation depends on assigned model work; the slider limits memory, not CPU/GPU activity. Keep this tab open to contribute.
Join to check GPU.
The app counts live GPU buffers or CPU arrays for weights, caches and working memory. Downloads, browser overhead, drivers and other tabs are excluded. Buffer limits do not measure free GPU memory. Each ready device holds a complete model. Chats use idle devices. Public chats stream to everyone viewing that conversation.
Waiting for a connection.