osFoundry / Platform
Hot replay & history
Every channel keeps the last N days of events. Reset a cursor, replay from any seq id.
What it is
Hot replay & history is osFoundry's per-channel event log. Every channel — the streams behind your chats, agents, and tools — retains the last N days of events, and each event carries a monotonic sequence id (seq id). Consumers read forward by tracking a cursor that points at the last seq id they processed. Because the recent window is kept hot, you don't have to treat a channel as a fire-and-forget stream: the history is there to be read again.
The core operation is simple. Move a cursor back to any seq id and replay from that point. An agent that crashed mid-run, a new consumer that just joined a channel, or a tool that processed events incorrectly can all re-read the exact same sequence rather than losing the gap or guessing what happened. Because osFoundry is a bring-your-own-key desktop workspace, this history lives with your workspace rather than on a vendor's servers — the event window is local-first, so replaying old events doesn't mean shipping your data back out to a third party. You reset the cursor, replay, and the consumer sees the same ordered events it would have seen live.
Why it matters: agent and multi-channel workflows are stateful, and state is only as trustworthy as your ability to reconstruct it. A retained, seq-addressed window turns "what did this channel actually see?" from a guess into a deterministic re-read. It makes debugging reproducible (replay the exact sequence that triggered a bug), makes recovery cheap (resume from the last good cursor instead of from zero), and makes onboarding a new consumer safe (backfill from a chosen seq id). The window is bounded to the last N days by design, so history stays useful and current without growing without limit.
Key capabilities
- Every channel retains the last N days of events, each stamped with a monotonic sequence id (seq id)
- Reset a consumer's cursor to any seq id and replay the exact ordered sequence of events from that point
- Recover crashed agents or interrupted runs by resuming from the last processed cursor instead of starting over
- Reproduce bugs deterministically by replaying the precise event sequence that triggered them, or backfill a newly added consumer from a chosen seq id
- Bounded, local-first retention keeps recent history hot on your device without unbounded growth or vendor round-trips
Use cases
- Agent developer: an agent crashes mid-run, so you rewind its cursor to the last committed seq id and replay the channel to resume exactly where it left off
- Workflow debugger: a downstream tool produced a wrong result, so you replay the same event range to reproduce and confirm the fix deterministically
- Integration builder: you add a new consumer to a busy channel and backfill it from a chosen seq id so it catches up on the last N days before going live
Frequently asked questions
How far back can I replay events in a channel?
Each channel keeps a rolling window of the last N days of events. Anything inside that window is addressable by its sequence id and can be replayed; events older than the retention window age out, so hot replay is designed for recent history and recovery, not permanent archival.
What does resetting a cursor actually do?
A cursor marks the last seq id a consumer has processed. Resetting it to an earlier seq id tells that consumer to re-read the channel from that point forward, receiving the same ordered events again. It doesn't delete or rewrite the events — it just changes where reading resumes.
Does replaying my channel history send data to osFoundry's servers?
No. osFoundry is a bring-your-own-key, local-first desktop workspace, and the event window is retained with your workspace. Replaying re-reads events that are already kept locally, so recovering or reprocessing history doesn't require shipping your channel data out to a third-party service.