Fleet view, not tab hunt
Every run in flight across the team, one screen: what's live, what's stalled, what needs a human. The morning check takes two minutes.
For engineering teams
Your coding agents, on one table
Every run visible. Every action attributed. Every hand-off explicit. Handbell is the shared surface over the agent fleet you already run: Claude Code, Cursor, Codex, your own harness.
The shape of the problem
Five agents, four repos, one person watching
Coding agents got good, so you run more of them. Now the constraint isn't generation. It's supervision. Whoever kicked off the run reviews the run, and everyone else finds out at the PR.
Second pair of eyes, live
Your teammate drops into your agent's refactor while it runs, reads the log, and flags the schema risk before the diff is 4,000 lines. Review spread across the team, not stacked on one person.
Hand off across timezones
Pass the run to the teammate coming online in Berlin. They accept with full context: plan, progress, open questions. The overnight run doesn't wait for your morning.
Built for how agents actually fail
Honest state, or no state
Handbell shows progress, not connection. A run that's burning tokens without advancing shows as stalled, because that's the state you most need to see. Run states come from what the agent actually does, not what it claims.
Steering mid-run is a first-class event, not a chat message the agent might notice. And when a run goes sideways, anyone who owns it can stop it cleanly at a checkpoint, without touching anyone's undo history.
Works with your stack. Handbell wraps runs from any harness that emits session events. Your agents, your runners, your repos. Handbell is the supervision layer, not another agent.
a table, mid-afternoon
run #217 · api-pagination · live · steering: sam
run #218 · flaky-test-hunt · live · 2 watching
run #214 · dep-upgrade · needs review
run #209 · perf-audit · stalled 22m · stop?
run #201 · docs-sweep · done 13:02 · shipped