Platform
Agents
Agents are the unit of work and the unit of trust. Mode is per agent, not per account.
Modes
Each agent carries either full_auto or approve_before_send. Making this per-agent rather than an account-wide switch is what lets a proven motion run hands-off while a new one stays on approval until it earns trust. Flipping the whole account to approval because one agent is new would cost the speed the product exists for.
The one human checkpoint
One preview: the ICP and the first three messages. After confirmation, agents send without per-message approval. Anything that adds a step before the first send needs a written decision, because time-to-value is a product constraint here rather than a slogan.
What stops an agent
- A reply. Any inbound message skips every pending queue row for that contact.
- An opt-out in prose. A keyword pass runs before classification, because a missed opt-out is a regulatory failure while a missed nuance is only a wrong badge in the inbox.
- The allowance. Overage is opt-in. By default agents pause at the plan allowance and nothing is invoiced.
- The daily cap. The queue filler enforces it before the launcher ever sees a row.