Changelog
Including the dull entries, the ones that removed a feature, and the one where we found our own cost reporting was wrong by roughly fifty times. A changelog that only carries good news is a marketing calendar with a monospace font.
29 August 2026
The four objects an outbound motion is made of finally live in one place. A template is the step graph, a sequence is one contact's run through it, an audience is a list, and a campaign is audiences times one template times senders. Plus the goal, the entry rule, the priority and the review mode. The entry rule is the point: what happens when a lead gets high intent was hard-coded in three places and editable in none.
29 August 2026
Autopilot sends eligible campaign steps automatically. Supervised mode holds steps for review. Both paths use the same pre-send checks; approval does not bypass suppression, budgets or sender readiness.
28 August 2026
Sending domains, sender addresses, warmup and inbound replies all run on one stack, and mailbox connects left the product entirely. A sender is now a from-address with its own connection row, so rotation, pacing and the age ramp apply per address. Replies arrive over an inbound webhook and run the same channel-agnostic pipeline a LinkedIn message does.
28 August 2026
Warmup-enabled senders exchange AI-written, deliberately mundane traffic between your own domains under the same ramp ceiling. Real sends, metered, logged, never shown in the inbox. Deliverability is scored per sender from delivery webhooks, and a bounce or complaint spike re-anchors the ramp automatically. What it does not claim: reputation at Gmail and Outlook is built by mail landing there, which a domain-only stack cannot fake.
27 August 2026
The internal cost reporting was wrong by roughly fifty times, and the reason was structural: it computed vendor cost as credits times a constant, which restates the price list and agrees with itself however wrong it is. The constant is deleted rather than corrected. Cost now comes from the provider's own response, from the vendor's reporting API, or from an operator-entered subscription, and where none of those has a number, it renders as unknown rather than as zero.
26 August 2026
Both sourcing paths now reveal before the write. A corpus pull reveals every candidate before the import; the LinkedIn sweep reveals every gate-passed lead by profile URL before the contact insert. Only people an address came back for become contacts, so the list never holds anyone you cannot write to. The manual enrichment controls came out with it. There is nothing left for them to do.
26 August 2026
Prospect-corpus contacts had no signal source at all and sat at heat 1 until a LinkedIn sweep happened to find the same person. The largest gap in intent, and the one that made the flame look broken. The watch now observes what a known contact recently reacted to, matches it against your monitored keywords, and writes it as evidence with the reaction's own timestamp. Budgeted at five profiles a tick and fifty per organisation a day.
25 August 2026
One entry now covers the whole chain: email first, then a few hours later the sequence moves to LinkedIn. The contact's latest post gets a like, no follow, and the connection request goes out with a fresh note. On accept, the first message waits 24 to 30 hours and is written fresh from their profile. Without a LinkedIn side, the email cadence runs unchanged.
24 August 2026
An event that could not reach a CRM used to mark the connector degraded and move on. It is now queued and retried with backoff. Five minutes doubling to six hours, dead after eight attempts, twenty-five per pass as tail work on the reply poller's tick. A reply reaching your CRM is not something to lose quietly.
22 August 2026
New connections default to ramping, and the launcher holds a young account to an age-banded schedule: five LinkedIn invitations a day in week one, ten to day thirty, twenty to day sixty, then your own limit. Sending domains ramp too. It is a per-connection switch and there is an attestation escape hatch for genuinely old accounts, but the default is now the safe one, because the consequence of a breach lands on the customer's account.
Before this
Rather than a change to it, so listing it would be a commit log with adjectives. The product went from nothing to the whole motion in a few months, and the entries above are where it started being worth telling you about.
Describe the company, confirm the keywords, connect a channel. The agents take it from there, and Supervised mode is available when you want to review before sending.
Start with sales. $2,000 or $5,000 onboarding, then at least $250 monthly credit funding.