What’s the intended setup for heartbeats

I’ve been trying to get a coherent setup out of OpenClaw but keep hitting issues with the way sessions work. From the docs I can’t tell what the intended setup is supposed to be, so figured I’d ask what everyone else is doing. Like most of us, I want an always on persistent agent that does stuff proactively. The persistent and proactive part relies mainly on the heartbeat to periodically wake the agent and initiate outbound actions without me prodding it. But what I don’t get is how the heartbeat should be configured - should your DMs and heartbeat share one main session? Or should heartbeats be in their own session? Both pose issues. Combined seems simpler, but compounds context throughout the day and pushes intermediate heartbeat thinking or inner monologue work out to your comms channel. Isolated fragments the context and the whole agent falls apart. Heartbeat wakes up isolated, injects a DM into the other session, I respond to the DM session which has no context on what the heartbeat was thinking. Then the next session executes with no context on what we discussed. Isolation seems to assume aggressive memory writes to the daily jsonl to provide cross session context, but the out of the box setup doesn’t prompt the agent to actually do that. But even if it did, memory is injected at the top of the session, so by the end of the day your cross session context is buried by session activity. To me, it seems like a better approach would be to have sessions represent a semantically contiguous block of activity. A conversation or agent work run with maybe a 30 min timeout. When sessions end they are summarized into the daily jsonl so that the next session knows what’s up but isn’t context flooded. Then you point heartbeat to the same place as DMs. Curious what other people are doing or what I am missing here!! Thanks in advance