Changelog

New updates and improvements to Starlight.

July 30, 2026

Ask the AI to book a meeting

The AI chat now schedules meetings end to end. Mention a contact with @ and name a time — "Schedule a meeting with @Ben Schouten next Friday at 1PM" — and it checks your real availability, books the slot, creates the calendar event with a meeting link, and emails the invitation in the contact's own timezone. If the time isn't open it offers the nearest open slots instead of guessing, and it can move an existing meeting to a new time on request. When no calendar or email account is connected, it says plainly what didn't go out rather than claiming an invite was sent.

July 29, 2026

Social listening was adding the people behind posts to Contacts on the strength of an AI guess about who they were — which meant rows with a name, a title, a company, and nothing else: no email, no industry, no location, no scores. A poster is now only added once we can independently confirm the person at that LinkedIn profile, and confirming them also fills in their industry, location, company size, work history, timezone, and match scores. The identity check compares names too, so a profile that resolves to a different person is left alone rather than quietly written onto the contact. People we can't confirm — and posts whose source didn't name an author at all — still show up on the Social page and can still be drafted to with one click; they just don't land in Contacts on their own. Duplicate rows for the same poster are prevented going forward.

Draft outreach no longer adds unverified people to Contacts

Drafting for a lead has always added them to Contacts, which made sense for people you searched for and chose. Applied to social posters — where a lead row exists for every post we find, verified or not — it filled Contacts with identities nobody had confirmed. Drafting now only auto-adds a socially-sourced person once they're verified. Nothing changes for leads from Search, Local, or an import.

Booked meetings show up on your Calendar

A meeting booked through your Starlight booking page was only visible under Scheduling → Bookings. The Calendar showed meetings from your connected Google or Outlook account, and only the ones with a saved contact on the invite — so a booking from someone new to you was filtered out of the very page meant to show your week, and a booking made before you connected any calendar never appeared at all. Both now show. Cancelling or rescheduling still lives on the Bookings tab, where the invitee gets emailed and the slot is released; deleting the meeting from your calendar now also cancels the booking instead of leaving it on the books.

Meeting prep on demand, and for the people who book you

AI meeting prep only ever ran at 7 AM, and only for meetings where someone on the invite was already in your contacts — which excluded the entire category it should have covered best: a meeting booked by someone new. Anyone who books through your booking page is now added to Contacts automatically, so their meeting gets prepped like any other. And the prep sheet has a Generate prep button, so you can pull it up the moment the meeting lands instead of waiting for the morning run. Same 5 blocks either way. The sheet also shows what the calendar actually holds — the agenda or description, and the location — whether or not AI prep exists.

Reschedule a booked meeting from the Calendar

Editing a meeting that came from your booking page used to change your calendar event and nothing else, so the booking behind it kept the old time — still shown under Bookings, still holding that slot as busy. Editing one now reschedules the booking properly: the guest is emailed, the old time goes back into your availability, and the calendar event is rebuilt at the new time. The new time still has to fall inside your availability, so a quick drag can't quietly double-book you. Cancelling from the Calendar goes through the same path.

Attendees resolve to your contacts again

On a busy month the Calendar asked for every attendee at once, hit a 50-address limit, and quietly got nothing back — so every guest rendered as a bare email address, none of them opened a contact profile, and the prep sheet showed the meeting as being with whoever the calendar happened to list first, which was usually you. It now asks in batches.

Meetings look the same everywhere

The meeting cards on Scheduling → Bookings and the Calendar agenda were built separately and looked it. They're now the same card: same layout, same duration and status badges, same Join button, same hover treatment. Bookings keep what's specific to them — the invitee, their notes, and Cancel or Mark no-show.

July 27, 2026

Your prompt now tells you when it's working against you

The prompt you write in the Outbound builder outranks everything else Starlight knows about good cold email — which is the point, but it means a reasonable-sounding instruction can quietly produce worse messages. Ask to "open with a compliment" and you tend to get praise about the recipient's job history, which reads like a database lookup. Ask for "no pressure" and those words usually land in the email, where they undercut the ask you just made. Say "mention we're an AI SDR" and you get a category label nobody can picture. Starlight now spots these as you type and suggests the instruction that gets what you actually meant — inline, dismissible, and never blocking anything. "Enhance with AI" fixes them too, rewriting the intent instead of just tidying the wording.

The quality reviewer got a lot harder to please

Every draft is scored by a separate reviewer before it reaches you, but that reviewer was almost entirely focused on whether claims were factually grounded — so an email could be accurate, follow your prompt exactly, be completely forgettable, and still score in the 90s. It now also asks the questions a reader would: does the observation about this person actually lead anywhere, or is it just a compliment? Is there exactly one ask? Would this email read identically if you sent it to anyone else with the same job title? Being accurate is treated as the floor rather than the bar. It also flags a draft that asks for a call when your prompt only asked to share something — overshooting your instructions, not just ignoring them.

Signals routes to the right person, or nobody

A signal at a company you're watching used to be sent to whoever matched your saved buyer persona, and if nobody really fit, it fell back to the most senior contact on file. That put company-wide news in front of people with no way to act on it. Starlight now works out who a specific event makes newly relevant — which isn't always your usual buyer — and if no one at the company genuinely fits, it skips the signal instead of drafting to the wrong person. Two related fixes: a recent job change is no longer described as more recent than Starlight can actually prove, and the event that triggered the outreach can no longer go missing from the message when a contact gets enriched mid-draft.

LinkedIn hides most post bodies from search engines, so social listening there was working from snippets. Now, when a watch scans, Starlight can also ask your Chrome extension to quietly open LinkedIn and X search in a background tab — as you, logged in — and read the full results: complete post text, engagement counts, and real profile links. The extra finds go through the same relevance judgment and land in the same Social feed. The extension only reads: it never posts, likes, comments, or connects, and while it does use LinkedIn's "Copy link to post" menu item to get each post's address, it catches the link in transit so your clipboard is never overwritten. Runs at a human browsing pace, costs no extra blocks, and can be turned off in Social settings → "Scan with your Chrome extension". Requires extension v1.5.0.

Suggested comments are upfront about who you are

Comments used to be forbidden from mentioning your product at all, which meant the only way to be useful was to sound like a neutral peer who happened to have opinions about tooling — exactly the posture communities punish once they find out. Now, when what you sell genuinely fits what someone wrote, the comment says so and says you work there: "full disclosure, I work at [your company]," then a plain reason it fits and your honest experience, including where it isn't the right answer. On a post asking the room what to use, the disclosure leads. On an announcement, the congratulation leads and the mention follows if it fits. When your product isn't relevant to the post, the comment doesn't mention it at all — a forced plug costs more than it earns. Everything it claims comes from your product description in Settings → AI, so it can't invent features or results, and it never uses links or marketing language.

Starlight was only ever listening for people with a problem, which meant it walked straight past some of the best openings there are — like someone announcing they've just joined a fast-growing company to lead its sales team. That person owns a number and hasn't picked their tools yet, and a congratulation that actually lands is the easiest first touch in sales. Social now watches for those moments too: new roles and promotions, funding rounds and launches, teams being built, and open-ended "what is everyone using for this?" questions — alongside the pain points and competitor complaints it already found. Suggested comments adapt: on an announcement, Starlight opens by congratulating the specific thing that happened rather than implying something's wrong, and auto-drafted outreach does the same.

Star thinks through your watchlist before writing it

The planner behind the Social watchlist now works through who your buyers are and which moments in their week create an opening before it proposes anything, and it deliberately spreads its angles across problems and moments instead of stacking up variations on one idea. Each watch also carries the actual phrasings people type in their posts, and several are searched every scan rather than one — the single biggest reason watches used to come back empty. A new Regenerate button on the Watchlist tab replans the whole set from scratch when it's gone quiet; watches you added yourself are kept, and finds already in your feed stay put.

The Add a watch dialog on the Social page no longer opens with a blank text box. Pick the kind of conversation to listen for: Competitor, Buying intent, Influencer, or Topic. Then type one thing, like a competitor's name, a problem your buyers post about, or an X handle, and Starlight writes the listening brief for you the same way its planner does. The dialog also suggests ready-made watches built from your profile's competitors and trigger events, each one click to add. You can still edit the name and brief before you start listening, and the Topic type keeps the original write-it-yourself flow.

July 26, 2026

The scanners behind the Social page were rebuilt end to end. Reddit — where buying-intent asks actually live — is now searched with a true recency bound, so fresh threads from the past week surface instead of being lost behind stale results. LinkedIn public posts come from a dedicated post-search source with real bodies and dates (no LinkedIn login or cookies involved — ever). On X, every find is now checked against the posts the search actually returned — a link that wasn't in the results never reaches your feed. A scan that finds several posts no longer loses them all because one was malformed, and if a scan fails entirely, its blocks are refunded automatically and the scan retries within hours. A social listening scan is now 4 blocks.

A new page in the sidebar's Intelligence group listens to public conversations on X, Reddit, and LinkedIn for the moments that matter — someone complaining about a tool you replace, evaluating your category, asking for recommendations, or rebuilding a team. There's no setup: Star turns your product, ICP, and competitors from Settings → AI into listening angles and scans daily. Every find is a real post, quoted verbatim with its link, and arrives with a suggested comment written to best practices — Copy & open post puts it on your clipboard and opens the post so you paste and reply yourself; Starlight never posts for you. Strong finds are resolved to the real person behind the post, added to Contacts, and (with Auto-draft on) turned into a reviewed-before-sending outreach sequence that joins the conversation like a peer.

July 25, 2026

Write your company profile the way your business actually works

The Details section in Settings → AI had about thirty fixed fields, and every business got the same thirty. Software with no integrations still stared at an Integrations box; a fact that mattered to you had nowhere to go if it didn't match a slot. It's now one fact sheet per card — a single document you write in, with a formatting toolbar, live preview, and suggested headings you can rename, reorder, or delete. Everything you'd already entered was converted for you, so nothing was lost. Lines you start with ">" are private notes to yourself and never reach the AI. The rules haven't changed: your fact sheet is still the only source of statistics your outreach can cite, and disqualifiers still cap a matching lead below 50.

Credits are now Blocks

The unit Starlight bills usage in has a new name: blocks. Nothing about pricing, allowances, or costs changed — a search still costs 12, drafting still costs 4 per lead, and your plan's monthly allowance is identical. Blocks better reflect what you're actually spending: units of computing that get consumed as Starlight works for you. You'll see the new name across the app, in Settings → Blocks (your usage ledger), and in the docs.

Scan with AI captures what outbound actually needs

Scanning your website now pulls far more than two paragraphs. Starlight researches your site and extracts verified proof points with their sources, capabilities, differentiators, integrations, compliance certifications, pricing model, and the objections buyers raise — plus, on the customer-profile side, buyer personas with exact job titles and seniority, company size and revenue bands, geographies, funding stages, trigger events, what prospects use today, and disqualifiers. All of it lands in the fact sheet on each card in Settings → AI, where it's yours to edit.

Proof points are the only statistics your outreach can cite

The draft-quality reviewer rejects any number it cannot trace, which used to mean real figures got flagged alongside invented ones. Your proof-point list is now that source of truth: a claim on the list is pre-approved and can be used verbatim, and anything missing from it is still treated as fabrication. Disqualifiers work the same way on the sourcing side — a lead matching one is scored below 50 regardless of how well it fits otherwise.

One scan fills both cards

The product pitch and ideal-customer-profile cards used to each run their own website scan, charging you twice for a single call's worth of research. Now one scan fills both, at half the cost. Also fixed: the Website field in Settings → AI silently discarded your edits when you saved, and on a brand-new account the details found during onboarding weren't being kept.

July 24, 2026

LinkedIn conversations read in the right order

The Inbox timeline now places each LinkedIn message by when it was sent rather than when it was drafted, so a connection note and its post-acceptance follow-up no longer stack together. The acceptance itself renders as a quiet event line between them — connection note, then "they accepted your connection request", then your follow-up — which is what actually happened.

@-mention contacts and companies in AI chat

Type @ in the Agents chat to search your Contacts and Companies and tag the exact record you mean. The pick becomes a chip in the composer, and the agent receives the record itself — so "make the @Ron Gabrisko messages more concise" edits that person's drafts immediately, with no name lookup and no ambiguity when two contacts share a name.

Company logo badge on your signature headshot

Your email signature's headshot now carries a small rounded-square company logo on its bottom-right corner — subtle branding that shows who you're with at a glance. The logo is pulled automatically from your Company URL, and you can upload your own in Settings → Signature to override it. The badge is baked into the headshot image itself, so it renders identically in every email client.

July 23, 2026

Stale connection requests retire automatically

LinkedIn connection requests that go unaccepted for 21 days no longer clog the Scheduled tab forever. Starlight retires them automatically — the held follow-up is cancelled, you get a notification with a link to LinkedIn's Sent Invitations page so you can withdraw the invites (they count against your invite limit), and each retired prospect appears in a new Expired invites section on the Tasks page suggesting an email touch instead. The contact stays in your CRM: an unanswered invite is a weak signal, not a no.

Signals route to the right person

Signal outreach now targets the person at the company who best fits each event — a security leader for a security acquisition, finance leadership for a funding round — instead of always drafting to the same saved contact. Saved contacts are considered first; when none fits, Starlight runs one company-pinned search, adds the best-fit person to Contacts, and drafts to them. Multiple signals at the same company now fan out to different people, and nobody gets a second signal draft within two weeks.

July 22, 2026

Signals: reach out when the timing is right

A new Signals page in the CRM section watches the companies in your pipeline for buying signals — funding rounds, hiring surges, job changes, product launches, and company news. Star picks what to track from your ICP and pipeline automatically, scans weekly, and can draft outreach that opens with the event (pending your review in Messages, as always). An optional Autopilot setting sends quality-judge-passing signal drafts without review. Discovery sweeps also surface net-new companies matching your ICP with a fresh trigger.

July 21, 2026

The closing line is yours to edit

The closing salutation above your signature — "Best,", "thanks", "cheers," — is now editable in the Outbound builder preview, just like the subject and body. Clear it entirely and the message closes straight into your signature. Whatever the preview shows is what the recipient receives.

Edit drafts directly in the preview

Previewed drafts in the Outbound builder are now editable in place — click a subject or body and type. Your edits are what gets saved on approve, with no regeneration and no extra credits. Edited messages are labelled "edited by you" instead of carrying the AI reviewer's score, and changing a setting that would regenerate the drafts now asks before discarding your edits.

July 20, 2026

New About and comparison pages

The website now has an About page covering who builds Starlight, plus honest feature-by-feature comparisons against Apollo.io, Amplemarket, Artisan, and 11x at /compare — including where each alternative is genuinely strong.

July 16, 2026

The AI agent can check your availability and book meetings

Ask the agent "what's open on my calendar Thursday?" and it reads your real availability — schedule, connected calendar, buffers, and booking caps. Tell it to book a time with a contact and it creates the booking, adds the calendar event with a meeting link, sends confirmation emails to both sides, and links the meeting to the contact. It always confirms the guest and time with you first, and the same abilities are available to external AI tools connected over MCP.

Recover meetings that fall through

When a prospect cancels a booked meeting or doesn't show up, Starlight now helps you win it back. Past meetings get a Mark no-show action in Scheduling → Bookings, and a lost meeting sends you a notification plus a new "Meeting fell through" section on Tasks that lists everyone worth rebooking — dropping them automatically once a new meeting is confirmed or a follow-up is queued.

Daily and weekly booking caps

Each event type can now cap how many meetings can be booked per day and per Mon–Sun week. Once a cap is hit, the remaining slots for that day or week disappear from your booking page. Your own Starlight bookings now also block their time slots even when no external calendar is connected, so back-to-back double-bookings can't happen.

Embed your booking page on any website

Scheduling → Page settings now provides a copy-paste embed snippet for each event type, so visitors can pick a time and book directly on your own site.

July 15, 2026

Local search results get a cleaner list and a detail panel

Local business results now use the same clean card layout as people search: business name, category, and address up top, with open/closed status, rating, and match tags as a single row of pills. Everything else — the owner's name and email, phone number, website, full weekly opening hours, and a link to the business on Google Maps — moved into a detail panel that slides in when you click a business, just like the profile panel in people search, complete with the same Notes and Activity tabs and arrow-key navigation between businesses.

The local search map now outlines the searched area

Local business searches now trace a dotted line around the area you searched — the actual city limits, following the real boundary rather than a plain box — so you can see at a glance which results fall inside the place you asked for and which spill into neighboring towns. It appears automatically on every local search, including ones you ran before this update, and needs no setup.

Local search results are now ranked by fit

Local business searches get the same intelligence as people search. Qualifiers in your query — "independent coffee shops, not chains, with outdated websites" — now become editable criteria chips in the search preview instead of being ignored, and every business is scored against your query, your criteria, and your Ideal Customer Profile. Results arrive sorted best-fit first with a match percentage and reason tags on each card (hover a tag to see why), strong matches get a highlighted ring on their map pin, and a small toggle above the list switches between best-fit and the original map order. Fit ranking is included in the existing search price — no extra credits.

A better-looking local search map

The map on local business searches got a full visual refresh. Every business now shows as a white pin with an icon for what it is — a cup for coffee shops, a wrench for plumbers, a fork for restaurants — instead of an anonymous dot, and nearby businesses group into clean white count pills that split apart as you zoom. Selecting a business highlights its pin in Starlight pink and opens a richer card with the business photo, rating, category, open/closed status, and address, and the map has new one-tap zoom controls. Hover a result in the list and its pin lights up on the map, exactly as before.

July 13, 2026

Booking pages follow your invitee's dark mode

Public booking pages now respect the visitor's system appearance. Invitees whose device is set to dark mode see the whole flow — meeting type list, calendar, time slots, confirmation form, and the confirm, cancel, and reschedule pages — in the same dark palette as the Starlight docs, while light-mode visitors see the page exactly as before. There's nothing to configure; it follows each visitor's own preference automatically.

Upload your own logo to your booking page

Scheduling → Page settings has a new Logo field. Until now your booking page borrowed whatever favicon it could find for your company domain, which is often low-resolution or simply the wrong mark. Now you can upload the logo you actually want (PNG, JPG, or WebP, up to 5MB) and it shows on your booking page, on each meeting type's page, and when someone reschedules. Nothing changes if you'd rather not: leave it empty and the company favicon stays the default, exactly as before.

July 11, 2026

Stop the agent mid-task

While the AI is working in the Agents panel, the send button now becomes a Stop button — or just press Esc. Stopping takes effect immediately: the agent halts, no further steps run, and no further credits are spent. A step already underway is allowed to finish rather than be cut off halfway (so a half-written change can't be left behind), and the conversation states plainly what happened — which steps completed, which was interrupted, and that nothing new was started. You can pick right back up with your next message.

July 10, 2026

Local business search

Search has a new Local tab for finding the businesses corporate contact databases don't index — plumbers, dentists, HVAC contractors, and every other local service. Type "plumbers in Austin TX" and results land on a live map next to the list: name, address, phone, website, rating, and opening hours, with the pins and cards staying in sync as you hover, click, and pan. Starlight then automatically researches each business to find the owner's name and email, so a map listing becomes a lead you can add to Contacts and draft messages for. Loading more results sweeps the area progressively, so dense metros go far beyond the first page.

July 9, 2026

A more polished email signature

Your signature sign-off got a quiet refresh. Your role and company now sit in subtle labeled chips beneath your name, and you can mask your headshot into a shape — circle, square, squircle, diamond, pentagon, or hexagon — from Settings → Signature, switching anytime without re-uploading. The sign-off also stays legible in dark-mode email clients now, and the redundant separate company-logo upload has been removed. Every change is styling only, so it has no effect on deliverability — no extra links or images are added.

July 6, 2026

See the AI at work: live agent cursors

Whenever Starlight's AI is doing something — reading your search, researching a lead, drafting messages, writing an inbox reply, or running tools from the Agents chat — a colored cursor with a name tag now glides across the screen to the spot where the work is happening, hovers there while it runs, and fades out the moment the result lands. Multiple actions each get their own cursor (up to five at once), colors stay consistent per task, and everything is purely visual: cursors never block clicks, pause when the tab is in the background, and respect your system's reduced-motion setting.

July 5, 2026

Inbox: one conversation per person

The Inbox now groups replies by person instead of by channel. If a prospect answers your email and then pings you on LinkedIn, that's one conversation — both messages appear in a single chronological thread with labeled Email and LinkedIn dividers, and small channel icons on each row show where they've replied. The confusing All / Email / LinkedIn tabs above the conversation are gone; in their place, compact email and LinkedIn filter buttons sit next to All and Unread in the list, and when a prospect has replied on both channels a Reply via toggle above the composer picks which channel your reply goes out on — each channel keeps its own AI draft and sent status. Also fixed: LinkedIn replies could be silently hidden from the Inbox list by an email-only bounce filter; they now always show up.

July 4, 2026

The chat agent can now run your virtual SDRs

Ask the Agents panel chat to set up a virtual SDR and it will do the whole thing in conversation — propose a goal, drafting brief, weekly quota, and schedule, then create the agent once you confirm. It can also update, pause, resume, or delete your agents, kick off a run on the spot, and answer "what did my agent do last night?" with the actual run numbers. A clear ask is all it needs — request an Autopilot agent and it creates it that way in one step, telling you plainly that passing drafts will send without review; it only asks follow-ups when your request leaves out something material, and deleting an agent always requires your confirmation. The same capabilities are available to external AI tools connected over MCP.

The AI knows your setup without checking

The chat agent now sees your workspace readiness at a glance — whether email and calendar are connected, whether the Chrome extension is installed and recently active, whether your booking page is live, and how many agents are running — so instead of asking "want me to check if your email is connected?", it just knows and tells you what's missing before a campaign. The connector status report also covers the Chrome extension and your booking page now, alongside email, calendar, and CRM.

Team roles: Admin joins Owner and Member

Settings → Team now works like the rest of your SaaS stack. There's a new Admin role between Owner and Member: Admins can invite teammates, remove members, change roles, and edit company settings, while billing, plan changes, and company deletion stay with Owners. When you invite someone you now pick their role right next to the email field, pending invites show the role they'll get, and every member row has a menu for changing roles or removing someone — with guardrails: only Owners can grant or revoke ownership, nobody can change their own role, and a company always keeps at least one Owner.

Attio CRM integration

Attio is now a fully connected CRM alongside HubSpot and Salesforce. Head to Connectors and click Connect next to Attio to link your workspace, then import your Attio people from the Contacts page's Import dialog — company names come along automatically — or push leads the other way with Export to Attio, available for your whole list or just the rows you've selected. Exports match people by email in Attio, so re-exporting updates existing records instead of creating duplicates, and imports are deduplicated by Attio record ID and email. The Pipedrive and Outreach placeholder cards have been retired from Connectors.

Referral program: give Starlight, get 3,000 credits

Settings has a new Referrals page with your personal referral link. Share it with a friend or colleague — when they sign up with your link and start a paying account, you both earn 3,000 credits. The page tracks every person who joined through your link as they move from signed up to activated to rewarded, along with your total credits earned. Referral credits sit on top of your monthly allowance and never expire with the billing period: they stay until you spend them, and each payout shows up in your Credits history as a Referral reward.

July 3, 2026

Salesforce CRM integration

Salesforce joins HubSpot as a fully connected CRM. Head to Connectors and click Connect next to Salesforce to link your org, then import your Salesforce contacts from the Contacts page's Import dialog or push leads the other way with Export to Salesforce — available for your whole list or just the rows you've selected. Imports are deduplicated by Salesforce contact ID and email so nothing doubles up, and exports update contacts Starlight already knows about instead of creating copies. Disconnect any time from Connectors; your data in Starlight stays put.

Every preview is now quality-reviewed before you see it

When you generate a preview in the Outbound builder, an independent AI reviewer now scores every message before it reaches your screen — checking that claims are grounded in your brief and product (no invented statistics), the tone reads human rather than templated, and each touch in a sequence adds something new instead of restating an earlier one. Messages that fall short are automatically revised once, and each draft in the preview carries a quality badge: "ready" means it passed review, anything else is worth a read before you approve. The score travels with the draft into Messages. Previously this review only ran for drafts created by your AI agents; now the builder gets the same treatment. The recipient picker also got smoother: a checkmark marks recipients whose previews are done, a spinner marks ones still drafting, and changing any setting mid-generation can no longer leave a stale preview from the old settings on screen.

July 2, 2026

LinkedIn sends that can't complete now pause and tell you why

Previously, if the browser extension couldn't finish a LinkedIn send — the profile was unreachable, a message button never appeared — the draft silently went back in the queue and retried forever, wedging every later step for that contact behind it. Now a send that fails three times (or hits an unrecoverable problem) pauses itself: it appears in Messages under Scheduled with a red banner explaining exactly what went wrong, you get a notification, and a one-click Retry puts it back in the queue with a clean slate. Paused steps no longer block the rest of that contact's outreach. Likes also got sturdier in extension v1.4.2: the extension reads the reaction state from LinkedIn's live button, so a like that landed is never mistaken for a failure — and a retry can never un-like a post.

July 1, 2026

Ask your agent to add another message to an existing sequence

You can now tell your AI agent "add a final closing email to Matthew and Jorge, asking if they're still interested" and it will add that extra touch to their existing outreach — no need to redraft the whole sequence from scratch. The new message is written in the same voice as the earlier ones, threaded as a reply to the original email, and lands in your Drafts under that contact as PENDING_REVIEW. Nothing sends until you approve it. Previously the agent could only start a brand-new sequence or edit a draft already there; adding one more step to outreach that's already underway wasn't possible.

June 29, 2026

Re-engage opportunities before they go cold

Tasks has a new Stale opportunities tab. When you've reached out to a contact and they haven't replied in two weeks — and nothing else is queued to touch them — they surface here, one row per person, most overdue first, so a warm lead doesn't quietly slip away. Click a row to open their thread and send a fresh message; replies, contacts still mid-sequence, and anyone with a draft already waiting are left out so the list only shows people who genuinely need a nudge.

LinkedIn automation that heals itself when LinkedIn changes

LinkedIn frequently reshuffles the internal names of its buttons and text boxes, which used to silently break sending until we shipped a browser-extension update. The extension now recognizes the Connect note field, Send button, message and comment composers, and Like button by what they ARE (their role and label) rather than a brittle internal name — so when LinkedIn rearranges things, your approved messages keep going out without waiting on an update. Every send still passes a safety check that confirms the right recipient and your exact approved text are in place before anything is sent; if it can't, it stops rather than send the wrong thing. And the first time anyone's extension adapts to a new LinkedIn layout, that fix is shared so everyone else skips the relearning.

June 23, 2026

Revise a draft by chatting with your AI agent

The redraft button on a message now opens your AI agent in chat with a starter that names that draft. Describe what you'd like changed — "less salesy opener, mention their recent funding" — and the agent rewrites that specific message in place, leaving your other channels untouched. Keep refining in the same thread ("now make it shorter"), and nothing sends: the revised draft stays in your Drafts for review and approval. Revising this way is billed per agent action instead of a flat redraft fee.

Smarter draft quality scores for sequences and rewrites

The AI quality judge now reads your whole sequence at once instead of grading each message alone — so a later follow-up or a warm break-up close is no longer marked "too empty" or "no product mention" when an earlier email already carries the pitch. And when you ask your agent to rewrite a draft in a new direction (say, a softer, no-pitch tone), it records that intent and the judge grades against what you actually asked for rather than the original brief. Scores now reflect the message you meant to send.

Clearer exports when you import your network

Connecting LinkedIn, Google, Instagram, or X now shows an Open button that takes you straight to that platform's export page, right above the upload area. For LinkedIn, it calls out the easy-to-miss detail that you must choose "Download larger data archive" — the other option leaves your connections out.

June 22, 2026

Surface your warm connections in search

Import your connections from LinkedIn, Google, Instagram, and X, and Starlight flags anyone you already know in your search results — with an In your network badge, a filter to show only people you know, and higher ranking for warm leads. When you draft outreach to a connection, the AI knows about the relationship and writes a warmer, non-cold message. Find it under Connectors → Map your network. Your uploads are read in memory and never stored; only the connections you choose to share are saved, and disconnecting a source deletes them.

Invite your team during onboarding

Onboarding now has an Invite your team step near the end, so you can bring colleagues into the company you just created. Copy a shareable invite link — anyone with it can join your company — or enter teammates' emails and we'll send each of them an invite. The step is optional, so you can skip it and invite people later from Settings.

Agents go deeper before running dry

Autonomous agents now page deep into searches that are already working instead of reinventing a query every run, and they fan out across complementary angles only when one is exhausted. The result: a well-targeted agent keeps surfacing fresh, relevant prospects run after run — and won't pause itself unless it has genuinely worked through its whole ideal customer profile. Agents also skip anyone they've already contacted, so the weekly quota goes toward new people. Weekly lead quotas now scale much higher for teams that want serious volume.

Email sending scales with your connected mailboxes

Outbound now fans sends out fairly across every connected mailbox rather than through one shared queue, so adding accounts increases how much can go out — while each mailbox still respects its own daily warm-up limit and never gets sent in tight bursts. The more mailboxes you connect and warm up, the more relevant outreach you can deliver per day without hurting deliverability.

Personalize your background gradient

Settings → Profile now has a Personalization section where you can pick the color gradient that drifts softly behind the app. Choose from a row of presets — Spectrum, Sunset, Periwinkle, Ocean, and more — and the background recolors instantly. Your pick is saved to your account, so it follows you across every device.

A refreshed Chrome extension card

The extension's controls now appear as a clean card that floats right on the page instead of a cramped toolbar popup, and it opens from the Starlight icon on any tab. Autopilot got smoother too: it picks up any queued sends the moment you return to Starlight and spaces them out more naturally. The default flow is unchanged — Starlight prefills your draft and you press Enter, and Autopilot stays off until you turn it on.

June 21, 2026

Deliverability: see your SPF, DKIM, and DMARC status

The Deliverability overview now checks the three DNS records that decide whether providers trust your mail. Each sending domain shows a clear pass or action-needed status for SPF, DKIM, and DMARC, with a direct link to the right setup guide for Google Workspace or Microsoft 365 when something's missing. Personal mailboxes like @gmail.com are marked provider-managed, so you only see what you can actually fix.

June 19, 2026

Meet Star — a guided, voiced onboarding

Signing up now starts with Star, a friendly virtual sales rep. A glowing orb greets you, then settles at the top of the screen and walks you through each setup step out loud, with a live caption that highlights every word as it's spoken. Prefer to read? A volume button in the corner mutes the voice and remembers your choice — the captions stay either way.

Companies: ICP scores, a deal board, and TAM

The Companies page now has three views. Table adds a Score column that rates every company for fit against your Ideal Customer Profile (Ideal Fit through Poor Fit). Board is a drag-and-drop deal pipeline — Lead through Closed Won/Lost — where each company carries an editable opportunity value, columns total their opportunities, and everything sums into your Total Addressable Market. Score groups companies by fit tier, strongest first. Scoring and deal estimates run automatically, cost no credits, and re-run only when your ICP or product changes.

June 17, 2026

LinkedIn follow-ups schedule themselves after the connection request

When you send a LinkedIn connection request, its follow-up Message now moves to the Scheduled tab automatically instead of waiting in Drafts — no need to approve each one by hand. It still holds until the prospect accepts, then sends.

June 16, 2026

Chatting with the AI agent is now free

Talking with the agent no longer costs credits — ask questions and plan as much as you like. You're only charged when it takes an action: each tool call costs 1 credit plus that action's own cost. Longer, busier conversations cost a bit more; quick questions cost nothing.

June 9, 2026

Agents: AI that runs Starlight for you

The new Agents panel (press A) lets you chat with an AI that searches leads, judges their fit against your ICP, drafts personalized outreach, and — with your confirmation — sends it. Create virtual SDRs with a goal, quota, and schedule that work on your behalf; every run leaves a full audit timeline, and an optional Autopilot mode sends messages that pass an independent quality judge.