See exactly what your AI agents did — task by task

When a team runs AI coding tools at scale, the hardest question to answer is a simple one: what did the agents actually do? Each tool keeps its own local history, on its own machine, in its own format. There is no shared place to see that Claude Code finished a task on one box while Cursor failed on another.

VibeControls closes that gap with an AI tool-call activity feed. Every AI coding tool running on a registered agent emits events — task.started, task.completed, task.failed — tagged with the source tool (opencode, claude-code, gemini-cli, cursor, or any tool you wire up) and carrying the full JSON payload of what happened. Those events land in the workspace as first-class records you can query and watch.

Because the feed is a live subscription, activity shows up as it happens: open the app and a completion on a remote agent surfaces as a notification the moment the tool reports it. A failed task announces itself instead of hiding in a log on a machine nobody is looking at. You can filter by agent, by source tool, or by time window to answer "what ran overnight?" or "which tool keeps failing on the staging box?"

This is observability for AI work, and it is deliberately separate from the human audit log. The audit log answers "which person changed what"; the tool-call feed answers "which AI tool did what, on which agent, and did it succeed." Every event carries its operation context, so it ties back to the right workspace and project.

For an engineering manager standing up an AI-native team, this is the difference between trusting the fleet and guessing at it. You get a task-by-task record of agent work across every tool, retained on your terms — old events can be pruned on a schedule — and visible from one workspace instead of scattered across a dozen terminals.