Credits are separate from model tokens. See Credits for
what draws from that balance and what does not.
Gateway limits
These govern requests through api.nemu.cc.
A refusal names the window it hit and when it resets, and carries a
Retry-After header. The request body limit matters most for coding agents,
because they resend the whole conversation on every turn.
Configuration limits
Chat limits
Context limits
Context is included on every plan and never spends credits.Computer limits
Total disk is across every project you keep, and it is what decides whether you
can create another one, not what you are charged. Disk is never billed. Each
project also has its own fixed size, listed in
Templates.
Session length is how long a preview keeps running after the last time you or
the agent worked on it, not a cap on total run time, so a long build is never cut
off part way through. Visitors keep a preview awake but do not extend it. See
Computer.
Every project gets the same 2 GB machine on every paid plan, so there is nothing
to choose and no size to upgrade to. Hours of building is your credits divided by
a modelled building hour, which makes it an estimate rather than an allowance:
credits also pay for search, page reads and research.
The Computer is not part of the free plan. See Computer.
Features by plan
Priority queue puts your chat jobs ahead of free traffic when the platform
is busy. Bypass load shedding means your requests are still accepted when
the gateway starts refusing others to protect itself.
Metered rates
Charged against your credit balance, not invoiced separately.
The Computer is metered only while it is awake, and it sleeps three minutes after
the last thing that touched it. In practice an hour of actively building is about
€0.039, and an hour awake but building nothing is about €0.02. See
Credits.