Email work to your AI agent.
It remembers your instructions and context, works with your files, follows up when needed, and wakes itself up later when something still needs to get done.
Receive → wake → reason → act → persist → schedule → wake again
@ Email your agent like a colleague.
Your agent reads it with everything it knows about your work.
It uses tools: files, memory, email, scheduling.
Durable context is saved where a human can read it.
If something needs to happen later, it schedules itself.
Built on primitives, not workflows
- Your workspace — a private GitHub repository per client: instructions, memory, projects, files. Comprehensible to humans, auditable in git.
- Scheduling is reasoning — a scheduled action is a reason to wake the agent later, not a canned action. It re-reads current reality before it acts.
- Deterministic authorization — every consequential tool call passes server-side permission checks, independent of the model.
- Model-agnostic — all inference goes through OpenRouter; the model is configuration, not architecture.