What is captured
Sessions from the coding agents you have enabled (seeresumecontext agents) for the directory that has been inited.
Turns are captured whole: full tool-call arguments and full output, not
truncated snippets.
How often
A background daemon syncs roughly every 20 seconds while it is running.resumecontext init starts it, and every project command except
agents restarts it if it has stopped, so in normal use there is
nothing to run and nothing to remember.
To push immediately rather than waiting for the next tick:
What lags, and by how much
Grep, read and queries: no lag
Grep, read and queries: no lag
These work on the synced turns directly, so they see every turn the moment
it arrives.
Search: by keyword at once, by meaning shortly after
Search: by keyword at once, by meaning shortly after
Synced turns are indexed as they arrive, so search finds them by keyword
straight away. Matching by meaning follows once they are embedded in the
background, newest history first — seconds normally, a few minutes after a
very large first sync. Until then a search still answers and says how much
is left, and your agent can use grep and read for the rest.
The current conversation: not there at all
The current conversation: not there at all
The session your agent is in right now has not been synced yet. It cannot
answer “what did I just do” from the archive. That is the one thing the
archive does not have.
An offline machine: until it comes back
An offline machine: until it comes back
A machine that is off or has the daemon stopped syncs nothing until it runs
again.