We use our own systems as teaching material, with their sources and current limits available to inspect.
These are our projects. They are not client case studies or measured returns from advisory work.
Keep decisions, tasks and file evidence together.
A local core that flags changed evidence and work needing attention. The download makes no model calls. Personal inbox, calendar and desktop integrations are separate.
Read the release and sourcePick up long agent tasks from a checkpoint.
Queues work and saves checkpoints. Claude, Codex and local adapters are available. Automatic context rotation remains Claude-only. Successful real local-model completion is still unverified.
Read the preview and limitsA shared handover between Codex and Claude Code.
Records decisions, evidence and the next action. It does not transfer a running conversation or its permissions.
Read the preview and release notesBring a recurring task and the questions around it. We’ll work out a useful place to start.
Arrange an intro call