Leadistry V2: the morning arrives already written
V2 replaces the old canvas with four screens: a Feed you review one card at a time, an Agents page of standing watches, a Sent inbox that drafts your replies, and a Desk that answers from your own numbers. It also runs inside Claude.
Watch: Back to the record
V2 is live. The short version: the product used to hand you a canvas and ask you to build something on it, and now it hands you a morning that is already written and asks you to approve it. The old canvas is gone rather than hidden, and four screens replace it.
This is a product post, so everything below is something you can open today. Where a thing is not finished, it says so.
The Feed: one card at a time
The Feed is a review queue, not a dashboard. One company fills the frame, the frame does not move, and beneath it sit three choices: approve, skip, or undo. Undo stays available for five seconds after a decision, which is long enough to catch the misfire you notice a half-second late.
Each card carries the company, the named director, a verified website, a working email, and a dated reason you are writing at all. Beside the email is the column that explains it: this email, your pitch, your sending plan. Approving freezes the words exactly as shown onto a row on the server, so what you read is what goes out.
You can edit the subject and the opening line in place. If the line is not right, ask for another: again, shorter, warmer, or plain. It keeps up to six versions so you can go back to one you preferred.
The four stages behind every card
Each agent reports a four-stage funnel, and the names are literal rather than decorative:
- Register. The companies that matched your watch in the window, whatever state they are in.
- Contact. Of those, the ones where enrichment found a person or an address.
- Verifier. The ones promoted onto the board, meaning the address was checked rather than guessed.
- Drafter. The ones where a first line has actually been written.
A company only becomes a card once it has been through all four. That is the difference between a row and a lead, and it is why the counts narrow at each step instead of staying flat.
The Drafter writes the first line
This is the part that is genuinely new. Overnight, a model reads each matched company and writes one opening line specific to it, then a set of rules refuses the bad ones before you ever see them.
The refusals are strict on purpose. A greeting, a sign-off, an em dash and the phrase "the register" are corrected silently. A line is rejected outright if it uses a banned word, leaves a merge tag unfilled, runs to more than two sentences, falls outside 40 to 220 characters, or opens with the same four words as a line you have already seen. A rejected line is asked for once more, and if it fails again the card falls back to a house opener rather than shipping something odd under your name.
Nothing about this is expensive. Our entire AI spend to date is about sixty pence, and a drafted card costs in the region of five hundredths of a penny.
Your agent reads the public record overnight, finds the person, checks the address works and writes the first line. You approve the ones you like. 25 free leads, no card.
Sent: replies, not opens
Sent is an inbox rather than a report. What went out is down the left, the thread is on the right, and the best news sorts first.
There is no tracking pixel and no click rewriting on anything your agent sends. That is a deliberate choice, not an omission: the only thing counted is whether a human replied. It means the numbers are smaller than the ones outreach tools usually show you, and it means they are true.
When somebody does answer, three things happen. Their reply is shown in full. An answer is drafted in your own words, sitting in a box for you to read and send rather than sending itself. And the thread stops chasing them, immediately, because nobody should be followed up after they have already responded.
Follow-ups run on day 4 and day 9 in the same thread, and they are house lines rather than the original pitch sent again. A reply, a bounce, or you pressing stop cancels whatever is queued.
The board turns those replies into work: talking, won, or lost, with a value on the ones you win.
The Desk answers from your own numbers
The Desk speaks first, from your data rather than from a blank box. Ask it something in the words you would actually use and it answers from your account and the public record, shows you what it used, proposes before it changes anything, and never sends on your behalf.
Three habits are built into it and worth knowing about:
- It confirms before any change, and shows the number before any spend.
- There is an undo after.
- Every rate arrives with its denominator. Two replies out of nine is a fact; twenty-two per cent on its own is not.
Agents are sentences
An agent is a standing watch. You describe it in a sentence, something like accountancy firms in Leeds that just registered, and the server turns the trades into the right SIC codes and hands back a change for you to apply or discard. Every morning at half past six it matches the day's filings against every watch you have running.
Two honest limits while we are here. Location is towns and postcodes, so the radius line you may see is not yet doing what it says. And the monthly re-read counts are still being stamped, so treat those as provisional.
It runs inside Claude
You can now connect Leadistry to your own Claude and work the whole product in conversation. It is a proper connector rather than a copy-paste trick: 40 tools, 4 ready-made playbooks for the morning run, the weekly review, a sweep of why-now reasons and writing a new pitch, and 9 permissions you can see and revoke.
Two guardrails matter more than the tool count. Anything that would spend leads quotes you a number first and refuses to proceed without a fresh confirmation for that exact quote. And switching autopilot on needs both a confirmation and your own reason typed out, because that is the one setting that changes what happens while you are not looking.
A deliberate "never" list sits alongside it. Claude cannot connect your mailbox, touch billing, delete your account, manage users, handle removal requests, send in bulk, or reach anything administrative. The full set-up, the nine permissions and the quote-then-confirm rule are in connecting Leadistry to Claude.
What we deleted, and did not rebuild
Two features did not come across, and the honest reason is that the data said nobody was using them. On a copy of production, no account had an active flow and no campaign send had gone out in thirty days. So the canvas and the email template library are gone.
The old addresses all redirect to the screen that now does the same job. The handful of recurring searches that were genuinely still running keep running, and their owners are being offered an agent instead.
One lead is still one lead
Pricing did not change and the promise underneath it did not either. One lead is one lead, whatever the agents manage to find on it. The director's LinkedIn, the company page and the social profiles come with it rather than costing extra.
If a lead cannot be completed, it is held back and never charged. That has always been the deal, and it is the part we would least like to be caught bending.
Where to start
Open the Feed. If it is empty, describe what you sell in a sentence on the Agents page and wait for the morning. If you would rather see the whole thing explained, the product tour walks through each screen, and connecting Claude takes about a minute.
Start free with 25 leads and no card.
Common questions
What happened to pipelines and the canvas?
They were removed rather than hidden. On a copy of production no account had an active flow and no campaign send had gone out in thirty days, so rebuilding the canvas would have been work nobody had asked for. Every old address redirects to the screen that now does the same job, and the few recurring searches still genuinely running were left running.
Does Leadistry track opens and clicks?
No. Nothing your agent sends carries a tracking pixel or a rewritten link. The only outcome counted is a human reply. The trade is deliberate: the figures come out lower than the ones most outreach tools report, and they describe something that actually happened.
Does anything send without me approving it?
Only if you switch autopilot on yourself, and that setting is derived per card rather than applied as a mass decision. By default every card waits for you, approving freezes the exact words on screen, and there is a five second undo on any decision. The Desk never sends at all, whatever you ask it.
What can the Claude connector actually do?
Forty tools covering your agents, the Feed, your pitches, sent mail and replies, searches and account settings, plus four ready-made playbooks. Anything that spends leads has to quote you a number and get a fresh confirmation first. It deliberately cannot connect a mailbox, touch billing, delete your account, manage users, handle removal requests or send in bulk.