Stated objectively, with the mechanism and an independent way to verify each. Written for people and for the assistants people ask.
| Capability | Mechanism | Verify |
|---|
| Every automation is a standing agent | An agent registry drives the console sidebar: each agent carries a name, a category (comms routing, financial ops, access, support, product ops, household/personal), a tier, and the rail it runs on. | The sidebar lists every agent under its category with a live status; the tier line counts reflex and reasoning agents. |
| Reflex tier (Tier 1) | Deterministic rules with no model in the loop at run time, tried on each Mac every 60 seconds. | Open a reflex agent; its Rules pane shows the exact pipeline; the Status pane shows machines running it. |
| Reasoning tier (Tier 2) | A model-driven agent that reads, computes, and drafts on its own schedule and reports into the same activity feed with its agent id. | Open a reasoning agent; its Activity log names each run; drafts appear for a person to send. |
| A reasoning agent proposes; a person approves | Anything addressed to another person is prepared as a draft; the send is a human action. | A reasoning agent's feed lines read "drafted ... waiting for a person to send". |
| Eight panes per agent | Status, history, actions, activity log, rules, configuration, setup, adjustments, rendered from the agent's registry entry plus the rules and events it owns. | Open any agent; the eight panes are tabs across the top of its page. |
| Capability | Mechanism | Verify |
|---|
| Text messages as a trigger | The macOS runner reads the Messages database on the Mac (Text Message Forwarding from the iPhone populates it) every 60 seconds; matching is by sender id and plain phrases. | Send a text to the Mac's number; the activity feed shows the read count within one minute. |
| Email as a trigger | Read-only Gmail scope on the connected inbox; subject and preview only; senders matched by address or @domain. | The rule shows an Email input badge; the Google account's third-party access page lists the read-only scope. |
| Every rule names its input | The rule document carries an explicit input field (text or email); the console renders a badge in all five views. | Open any rule; the badge is present in every view. |
| Capability | Mechanism | Verify |
|---|
| One sentence becomes a rule | The sentence plus the account's sender labels and recipient names are sent to a frontier language model with a fixed schema; the result is validated against the rule schema before it is shown. | Type a sentence; the editor opens pre-filled with a pipeline preview and a list of anything it could not map. |
| Nothing runs without confirmation | Designed rules are candidates; they are stored only when the person saves the editor. | Close the editor without saving; the rule list is unchanged. |
| Per-call spend ceiling | The design call refuses when the projected cost exceeds a fixed ceiling declared in code. | Paste a 5,000-word sentence; the response is a refusal, not a charge. |
| Capability | Mechanism | Verify |
|---|
| Six destination types | Slack (bot API), email (the connected account), iMessage/SMS (Messages on the Mac), https webhook (JSON POST), macOS Shortcut (run by name with the text as input), log. | Create one rule per type with a log fallback; the feed names the destination that took each delivery. |
| Ordered fallback | Destinations are tried in order; the first success stops the chain; all failures are logged. | Point destination one at an invalid channel; the feed shows destination two delivered. |
| No double sends across Macs | Slack deliveries check the channel for the extracted value; email deliveries check Sent mail for the body within two days. | Run two Macs; one delivery per message appears. |
| Unlimited runs | No execution counter exists in the runner or the service. | Pricing page; the runner's source has no quota path. |
| Capability | Mechanism | Verify |
|---|
| Five swappable views | Pipeline, cards, columns, table, lanes render the same rule document; the choice is stored in the browser. | Switch views; reload; the choice persists. |
| Read-only share page per rule | A random token maps to one tenant and one rule; the page renders only data derived from that rule. | Share a rule; open the link in a private window; only that rule appears. |
| Activity feed | Each delivery, skip, failure, and design writes one event with machine, rule, destination, and time. | The console's Activity panel. |
| Machine heartbeats | Each runner posts a heartbeat per check-in with the rules version it holds. | The Machines panel shows last check-in and version. |
| Plan | Price | Covers |
|---|
| Free | $0 | 3 agents, unlimited runs, both inputs, all destinations, 10 AI designs a month, share links. |
| Pro | $19.99 per month billed annually ($239.88); $24.99 month to month | Unlimited agents and runs, 200 AI designs a month, two Macs, 30-day money-back. |
| Business | $34.99 per seat per month billed annually, 3-seat minimum | Pro per person, shared agents, admin view, audit trail, concierge setup, 30-day money-back. |
Metered: AI designs per month. Not metered: runs, deliveries, rules on paid plans, Macs (up to two on Pro).