How Revnu Works

Revnu is a growth agent. It finds buyers, writes and sends outreach, runs ads and writes the pages that get a business found in search and AI answers. The API does not give you those tools directly. It gives you the same controls a customer has in the dashboard: see what the agent wants to do, say yes or no, tell it what to do next, and read what happened.

Review cards

Anything that needs your say-so becomes a card in the review queue: an email draft, a blog post, a budget change, a question. A card has a kind, a sourceType naming the work, and a status of pending or needs_revision. The kinds that ask you to decide are approval (ship this), question (often with textInput) and choice (pick one of options). The others record the agent's own work: review (a judgment worth a second look), follow_up, setup, and update (a result with nothing to decide). Expect new kinds over time.

Each card lists the verbs it accepts right now in actions, and the label the dashboard would put on its main button in primaryLabel. Build your UI from those two fields rather than from sourceType, and new kinds of work render correctly without a code change. The verbs:

Verb	Does
approve	Ships the work: publishes, applies, sets up, or approves
answer	Answers a question card
choose	Picks one option on an option card
send	Sends a Gmail draft or reply, optionally edited
request_changes	Sends the card back to the agent with your feedback
snooze	Hides the card until a time
dismiss	Declines it; nothing ships

The trust ladder

Revnu tracks how you resolve each kind of work. A clean approval counts in full, an edited one counts half, a dismissal counts zero. Once a kind of work has enough volume and a high enough approval rate, Revnu asks whether it may ship that kind on its own. Only after you agree do its cards carry autoResolveAt, the time they ship unless you act first: at least four hours for sends, a day for reversible work such as blog drafts.

Some work never ships on its own: anything that spends money, and replies inside a live email conversation. Snoozing a card cancels its deadline, and dismissing or sending one back resets that kind's streak. Cards that wait more than seven days before their kind has earned trust expire, so the queue never turns into a backlog.

The agent

The agent runs in its own sandbox and wakes on a schedule, on new mail and replies, and when you message it. Send it a message the way you would in the dashboard chat, then wait for the reply inline or poll. Agent status says whether it is up, busy, and when it next wakes.

Workflows

A workflow is a recurring job written in plain words: a schedule such as "every weekday at 09:00 Europe/London" and a list of steps. A step can carry a gate, which makes it wait for a person before it acts. The agent reads the steps on every run, so volume ("five per day") is an instruction in a step, not a numeric field. See Create a workflow.

Results

Results is the scoreboard the agent reads when it wakes: one section per channel, each either available with numbers or unavailable with a reason. A channel that is not connected says so rather than reporting zero.

Leads

A lead is a company the agent has researched, with a status, a score, contacts, your own columns, and the draft it would send. You can read leads, set their labels, and keep a block list of companies, mailboxes and LinkedIn profiles that outreach must never touch.