A guided interview, about 25 minutes
It asks what you built, who it is for, and what you want. Out of that comes a working picture of your product, your personas, and where those people actually spend time.
ClawBot Crew · GTM Agent
Most growth tooling optimises for volume, which is why so much of it gets accounts banned and brands disliked. The GTM Agent starts in read-only: it learns your product, works out who it is for and where they already gather, and reads those places for a while. It does not post on your behalf until you explicitly tell it to move.
Watching is the default, not a setting you have to find. Everything it learns cites a source you can check.
Four things, in the order that keeps you welcome in the rooms you want to be in.
It asks what you built, who it is for, and what you want. Out of that comes a working picture of your product, your personas, and where those people actually spend time.
It reads the communities, learns their rules and their rhythm, and reports back. Nothing is posted in your name until you say to move forward — that is a hard gate, not a default you can trip over.
Every community, persona and contact it records comes with at least one source. "This subreddit bans self-promotion on weekdays" is a link, not a vibe.
When a DM is not viable it uses a per-persona Gmail, separate from your primary address, plain text only, with a daily cap. The limits exist so outreach stays something a human could plausibly have sent.
Slower at the start, which is the entire point.
About 25 minutes of plain questions. This is the only heavy step, and skipping it is how agents end up confidently pitching the wrong people.
YouWho your users are, which communities they are in, what each one allows, and what tends to work there. All of it filed with sources so you can push back on any single claim.
AgentFor a while it just watches: what gets asked, what gets removed, what the moderators are tired of.
You get findings, not activity. This is the stage most tools skip, and it is why they get their users banned.
AgentOnly then does it act — posting, replying, or reaching out by email within the caps. Contacts go into a pool with a do-not-contact list that cannot be reversed once someone is on it.
YouStart with the tutorial; everything else builds on it.
These are constraints by design. They are why it is safe to point at a real community.
Watching is the starting state. Nothing appears under your name in any community until you explicitly move it forward.
The do-not-contact list is immutable, and right-to-be-forgotten requests are honoured. Once a person is off the list of contactable people, no later instruction puts them back on.
Cold email is plain text, per-persona, and daily-capped. If your plan depends on thousands of sends, this is the wrong tool and it will not pretend otherwise.
It can tell you the rules, the timing and what has worked. Whether your product is worth their attention is a product question, and no amount of targeting fixes it.
One subscription per crew agent, with its own monthly credit allowance.
Includes 1,500 credits per month for research, community monitoring, the contact pool and outreach. Free trial available.
Crew agents are hired on top of a paid ClawBot plan. One base plan covers every crew agent you hire.
It is built to be the opposite, and the constraints are structural rather than advisory: watch-first by default, per-persona email with a daily cap, plain text only, and a do-not-contact list that cannot be undone.
If what you want is volume, this will frustrate you. That is the intended outcome.
It reads each community's rules and reports them before acting, and it re-checks for changes. It cannot guarantee a moderator's judgement, which is exactly why the posting gate is yours.
So cold outreach never runs from your primary address, and so reply handling and reputation stay contained. Plain text only, because plain text is what a person actually sends.
Yes — tag them in chat. It pairs naturally with the Content Agent for publishing and the Product Agent for turning what users say into a build plan.
Each crew agent is its own subscription, with its own memory and its own credits. They can hand work to each other when you tag them in chat.
Twenty-five minutes now is what stops it confidently pitching the wrong people for a month. Then let it watch for a few days before you let it move.