Bring your own Claude. We bring the leads.

“Morning. What's on my Feed today? One line each: who they are, why now, and whether there's a verified email. Don't approve anything yet.”

“Change my profile to target letting agents in Doncaster, then price a search against what's left on the account. Don't start anything until I've seen the number.”

It updates the profile, prices the search, and stops. Starting it is a second step that needs that quote's own token, so a search can never be started without being priced first.

“Anyone replied since Friday? Read me what they actually wrote, not a summary, and tell me which ones already have an answer drafted or sent.”

“What has my accounts-due agent turned up in the last fortnight, and where did each one get to?”

“Open my last search. How many leads did it deliver against what it cost, and list the ones in Bristol with the director's name.”

“Write me two pitches for my bookkeeping practice aimed at companies coming up to their first year end. Different angles, not different wording, and show me both before you create anything.”

It reads your profile and your results, shows both, and creates nothing until you choose one.

“The first line on the Warrington roofing card reads like a robot. Make it warmer, show me the old and new lines side by side, and don't approve it.”

“The kitchen fitter in Stockport has replied asking what we charge. Draft an answer in my voice that gets him talking rather than quoting a price, and don't send it until I've read the exact words.”

Sending an answer needs a confirmation from you, with the exact words you read in front of you.

“Run 'Cash flow' against 'Year end' as an A/B test on the same cards. Tell me roughly how many sends it needs before the verdict means anything, and don't send anything.”

“Which of my pitches is actually getting replies? Put the number of emails next to every rate, and if a slice is too small to judge, say so instead of ranking it.”

“Mark Hartley Joinery as won at £2,400 and tell me what that does to my cost per deal this month.”

“What's the A/B test saying? Don't guess: if it isn't settled, tell me how many more sends or replies it needs.”

“Go through every agent I'm running, check what each has raised against my results by reason, and tell me which to narrow, pause or replace. Propose only, I'll say yes per change.”

“Run my weekly review. What's talking, won and lost this period, what is the plan costing me per deal, and where would you point the next batch of leads? Quote it, don't start it.”

“Set up an agent watching for brand-new landscaping firms in Norwich. Watch only for now, and tell me what it would cost per day if I switched auto-enrich on later.”

“Do my morning run. Show me the subject and first line of each card, say which you'd approve and why, and I'll say yes per card. Leave autopilot alone.”

This is the morning routine by name. It proposes per card and approves only what you say yes to.

“Stop chasing the roofing firm in Halifax, they've asked to be left alone, and put them on my do-not-contact list.”

It stops the thread. The second half it cannot do: no connection can touch a do-not-contact list, and it says so rather than pretending.

“I'm away all next week. Switch autopilot on so the morning emails keep going, and tell me exactly what that changes before you do it.”

It explains what autopilot changes first, and needs your confirmation in your own words before it touches the switch.

“Send the 'Free audit' pitch to all 80 roofers from Tuesday's search tonight in one go so I don't have to go through the Feed.”

It declines. There is no bulk send. Cards are approved by name, each with the words you have seen, never as a list.