-
Port 143: three ways a mail client can lose your password before TLS starts, and the sales agent's refusals
The self-hosted sales agent now speaks IMAP STARTTLS on port 143 and refuses a stripped STARTTLS, PREAUTH or injected bytes before login, checked on Dovecot.
-
The sales agent meets a real mail server: 32 checks on Dovecot and Postfix, and the bug the fakes hid
The self-hosted sales agent passed 32 of 32 checks on real Dovecot and Postfix, after fixing a bug that would have emailed an answered prospect twice.
-
A wrong no is the one nobody reads: sorting an AI model's mistakes by what they cost
A local 7B model's 37 false stops in our sales agent, sorted by what each category does. 16 were leads closed unread; now 3.
-
Canada Day on a Saturday is a Monday off
Sales Agent's days off now include Canada's federal holidays. They match 85 of 85 dates on a second source; canada.ca itself still has to be checked.
-
A free 7B model reads sales replies better than our rules, and still can't be trusted with a stop
A free local 7B model reads unfamiliar sales replies far better than keyword rules, but mistakes live leads for opt-outs. How the agent now uses it.
-
We asked a free local model to check our sales agent's grading. It graded like the rules did.
Llama 3.1 8B, run locally with no key, labelled 298 real Enron replies. It agreed with our agent readers less than they agree, and leaned toward the rules.
-
The free page gets the call buttons too
Sales Agent's free page now marks each due call Spoke, Voicemail or No answer, re-queues misses, moves later steps and remembers it all in your browser.
-
Call outcomes without a server: one CSV row per dial
Sales Agent's CLI now reads a calls.csv: spoke or voicemail ends a call step, a no-answer call comes back the next business day, and call prep lists each try.
-
Your Sent folder knows which emails you sent. It has no idea you made the call.
Sales Agent's self-hosted server now has a Done button for calls, LinkedIn touches and letters in its daily queue, kept across restarts, with Undo.
-
Don't book the next call on their holiday: call prep now knows the prospect's days off
The sales agent's call brief now lists the prospect's next days off, and the follow-up email flags a next step dated on one.
-
When a cold call rings out, the rest of the cadence should wait
Sales Agent now moves a prospect's later steps back when a call is retried, and keeps retries off US holidays, so nobody gets a call and an email the same day.
-
A call that rang out is not a call you made
Sales Agent's Today's queue takes an outcome per call: no answer brings it back the next business day, up to three tries, and call prep lists each try.
-
Seven out of seven was luck: a pre-registered test on a local model fails
A 14B local model got 7 of 7 right on a small sample. We wrote the bar down before testing it on 611 more replies, and it got 2 of 7. Nothing shipped.
-
Twice the model, mostly the same mistakes: a 14B local model reads sales replies
A free 14B model on four CPU cores reads sales replies slightly better than a 7B, still reads live leads as opt-outs, and is no better as a second reader.
-
Good Friday is not a business day: UK bank holidays in the sales agent, checked against GOV.UK
Call retries in the sales agent now skip UK bank holidays. A test against GOV.UK's own file caught an Easter bug and a one-off holiday.
-
Texas works Columbus Day: 51 US state holiday calendars in the sales agent, checked against python-holidays
The sales agent's days-off list now takes all 50 states and DC. Its US federal rules, unchecked until now, match a second source too.
-
"We're not interested" stopped, "We are not interested" didn't: one contraction in a sales agent's rules
A guard written for other people's no also caught the writer's own. The fix was one lookbehind, checked on 76,565 real Enron replies.
-
Whose holiday is it? Canada's thirteen calendars and the prospect on the other end of the call
Call retries skipped the rep's holidays, not the prospect's. Canada's 13 provincial calendars as rules matched 855 of 855 published dates.
-
Every "I'm out of the office" our sales agent paused on real mail was typed by a person
Splitting auto-responder notices from people's own away replies cut the sales agent's wrong silent stops from 3.9 to 1.8 per 1,000 real replies.
-
"No. I need to know how many beds" is not a decline: a fifth round on real mail
Fifth round of sales-agent reply fixes, scored once on an unread Enron half: wrong stops 7.2 to 5.5 per 1,000 replies, one right stop lost.
-
The third round of fixes cut 52% of wrong stops on mail we'd read, and 23% on mail we hadn't
Third-round fixes to the sales agent's reply rules, scored once on 9,656 unread Enron replies: wrong stops 10.9 to 8.4 per 1,000, right stops 47 to 46.
-
"I'm out Friday" is not an away notice: a fourth round on real mail
Fourth round of sales-agent reply fixes, scored once on an unread Enron half: wrong stops 7.2 to 5.0 per 1,000 replies, one right stop lost.
-
Drop your mailbox export on the page: 32,211 real emails, and the bug they found
Sales Agent now opens .eml and .mbox exports in the browser, nothing uploaded. A 90 MB mbox of real Enron mail also turned up a crash and a bounce bug.
-
Our sales agent now checks your Sent folder before it tells you who to email
The self-hosted Sales Agent reads five header fields from your Sent folder, so today's queue drops emails you already sent. And IMAP refuses Node's "Sept".
-
Your sent folder already knows which follow-ups you made
Sales Agent now reads your sent mail from a Takeout or Thunderbird export, so follow-ups you already sent are not queued again. Checked on 7,836 real emails.
-
When the held-out half says stop: a rule round that halved errors and changed nothing
Round seven halved wrong stops on its tuning half, then moved one stop in 90 on unread Enron mail: 4.0 to 3.9 per 1,000. Why, and what comes next.
-
We fixed 140 wrong stops on mail we'd read. On mail we hadn't, it was 83.
Second-round fixes to the sales agent's reply rules, scored once on 9,584 unread Enron replies: wrong stops 17.8 to 9.2 per 1,000.
-
Two blind readers, one inbox: how much of a classifier's score is the grader
Two blind readers agreed on 81% of 107 real replies, and the reader moved the score as much as a round of fixes. Round six: 5.6 to 4.3 wrong stops per 1,000.
-
Our sales agent's second opinion was trained on 120 replies. We had 792.
The sales agent's word model now trains on 792 labelled replies. Scored set by set on data it never saw, it reads 28 of 57 hard replies right, up from 19.
-
The From line that splits your mailbox: reading real Gmail and Thunderbird exports
The sales agent's mail-file reader on eleven mbox files people published on GitHub: two corners were wrong, and Python's mailbox module gets both wrong too.
-
113 real people asking to be taken off a list: how many did the reply rules hear?
Sales Agent's reply rules caught 103 of 113 real opt-out requests in the Enron corpus. The ten misses, and the five patterns that fixed them.
-
Seven in ten: what a reply classifier gets wrong on 19,000 real emails
Sales Agent's reply rules on 19,153 real Enron replies: most stops were wrong. Fixes tuned on one half, scored once on the other.
-
Twenty-first reply set for the sales-agent: questions and objections mostly read right
Twenty fresh cold-email replies about pricing, rivals, security and booking a call: 17 right before any change, 19 after two narrow patches.
-
Our cold-email checker lost the Germany consent warning when you typed 'Germany'
The free pre-send checker only understood two-letter country codes, so 'Germany' or 'United States' silently skipped the consent rule. It now reads names.
-
A cold-email checker that said 'no opt-out' to a perfectly good opt-out
Trying realistic drafts through the free pre-send checker found four opt-out phrasings and German addresses it wrongly rejected.
-
A Draft Patch That Broke an Older Reply Set: the sales agent's fourteenth test
Fourteenth fresh set of 31 sales replies: 26 right before any change, 31 after. One draft patch cost an older set a point, so old sets stay in the loop.
-
The sales-agent now reads a sabbatical auto-reply as out of office
A 20-reply probe of auto-replies, bounces and non-English opt-outs found one miss, now fixed.
-
The Lead Who Said "Wrong Person" and Meant Yes: probing an AI sales agent for false stops
A 20-reply probe found the reply rules marking six live leads as departed contacts. Narrow vetoes took it from 9/20 to 18/20 and found a missing opt-out.
-
The polite no that sounds like a yes
A probe of an open-source sales reply agent with declines that sound positive: 13 of 20 right before, 17 of 20 after, and the one case still wrong.
-
When a cold-email reply mentions a bounce but is a person: a 17th test set
Twenty replies from people who mention bounces, trips and leave while asking a question. One narrow fix stops three live leads being marked bounced.
-
When 'not now' is really a question or a final no: an 18th test set
Twenty replies that sound like a pause or a handover but are a question, a price objection or a flat no. Three narrow fixes took the rules from 9 to 15 right.
-
When 'don't stop' gets read as 'stop': false opt-outs in a cold-email reply triager
A probe of 20 stop-word look-alikes made our open-source reply triager suppress 6 interested leads; a narrow veto fixed all 6.
-
Thirteenth reply set for the sales-agent triage: 18 of 30 before patching, 30 of 30 after
Thirty fresh cold-email replies scored once before any change: 18 right, two unsubscribes missed. What we patched and why 30 of 30 is not a forecast.
-
Twelfth reply set for the sales-agent triage: 25 of 30 before patching, 30 of 30 after
Thirty fresh cold-email replies in six languages scored once before any change: 25 right, one unsubscribe missed. What we patched and what we did not trust.
-
Clean your cold email list before the first send
A free, offline checker that sorts a prospect list into send, review and drop before you email anyone.
-
A one-character bug in a calendar file generator
The open-source sales agent writes a tentative .ics file when a prospect proposes a time. A semicolon in a name was not escaped.
-
The sales agent's approval queue now hands you the meeting as a calendar file
When a prospect replies 'Thursday at 2pm', the self-hosted sales agent offers a tentative .ics file beside the draft. It sends nothing and guesses no time zone.
-
From a cold-email reply to a calendar file, with no time zone guessed
The open-source sales agent writes a tentative .ics file for an exact meeting time a prospect proposes, and refuses on a part of the day.
-
Teaching a sales agent to read replies: eleventh test set, 26 of 30 to 29 of 30
A fresh set of 30 cold-email replies: the rules got 26 right before patching and 29 after, with no opt-out missed and no older set worse.
-
The February that had 31 days
A bug in the Sales Agent's date reader: an out-of-office reply saying 'back 31 February' was turned into March 3. Fixed, plus 'next Tuesday' is now understood.