Built in public.
Monthly roll-up of what shipped — features, infrastructure, and trade-offs. We'll edit this in plain English instead of release-notes-as-marketing-copy. If something here changed your workflow, reply to the launch email; we read every one.
- May 2026
Cold-Start-Inheritance — first-campaign seeding from opt-in projects
Third vector-similarity surface, opt-in only. When you create your first campaign on a fresh project, Overwise looks across opt-in customer projects with similar pitches, finds the top neighbors, mean-pools their converters weighted by source-similarity, and returns up to 50 cross-tenant lookalike leads as seed for your first campaign. The opt-in lives on the settings → compliance page with a reciprocity-arrow consent dialog: you contribute → you receive. Revoke anytime; revocation is soft (your embedding stays available for analytics, but neighbors stop seeing your converters).
- May 2026
Lookalike-Loop + Reverse-Discovery (vector-similarity leads)
Two new lead-finding modes that fire after you have ~5 converters in a campaign. Lookalike-Loop drafts an AI Inbox suggestion at every 5-converter milestone: 'here are 50 vector-similar leads we found — import them?' Reverse-Discovery is the user-driven mirror: paste 3-20 of your best customer domains, get 50 lookalike leads back in under 2 seconds, attach them to a new campaign at launch. Both are powered by an Atlas Vector Search index over the cross-tenant lead pool — recall hit 99.3% precision@9 on the launch eval.
- May 2026
Per-project subscriptions + tier-based feature gates
Billing rewritten to per-project subscriptions (Stripe-canonical). Three tiers — Starter $99, Growth $249, Founder Team $599. Trial unlocks all features for 14 days; post-trial gates fire on attempts to use higher-tier features. Lead-cap chip added to campaign-detail view. Overage opt-in at $0.10/lead, hard-capped at 2× tier limit.
- April 2026
Real-time hot-lead trigger scanner
Per-project hourly scanner watches your active leads for buying-intent signals: hiring spikes (Greenhouse + Lever public boards), funding announcements (SERP scan), tech-stack changes via Wayback CDX snapshots, and competitor churn. Hits surface as HOT_LEAD_BOOST tasks in the AI Inbox with the evidence inline — and the lead's outreach state jumps the queue. Free for the four trigger sources we host; Serper funding scans use shared API credit.
- April 2026
Product Knowledge surface + Discovery-only CSV export
Two trust-loop additions for users who aren't ready to autopilot yet. The Product Knowledge card on Settings holds your pitch, ICP details, language toggles, qualification threshold, and per-product absence probes — the agent reads it before every decision, so editing it changes outreach instantly without re-creating campaigns. And Discovery-only campaigns get a one-click CSV export of qualified leads, audit-logged with campaign + status + count, so the segment of users who want to QA the list before sending can do exactly that.
- April 2026
Lead-centric dashboard redesign
Campaign detail rebuilt around leads, not templates. Three tabs — Leads, Insights, Activity — replace the old configuration-first layout. Per-lead timeline shows the agent's decision log entry-by-entry: SEND / WAIT / HOLD / TERMINATE plus the rationale and cited signals. AI Inbox got a per-task-type renderer registry: APPROVE_OUTREACH, MANUAL_OUTREACH_DISPATCH, RESPOND_TO_REPLY, MESSAGE_VERIFICATION_FAILED, CAMPAIGN_AUTO_PAUSED, plus the May 2026 vector-similarity tasks. Brand Voice editor moved to Settings.
- April 2026
Agentic outreach loop replaces the fixed sequence
The fixed 5-step email sequence is gone. Replaced with a LeadOutreachAgent — a Sonnet executor that decides per-lead what to send and on which channel, with an Opus advisor reviewing edge cases. MessageVerifier (cite-or-discard) discards drafts whose claims can't be supported by real signals. Per-lead reasoning surfaced in the campaign detail timeline.
- March 2026
Source Library v2 + AI-driven source routing
Lead-discovery engine extended to 9 enabled sources: OpenStreetMap Overpass, Outscraper, Serper SERP, ProductHunt, GitHub, Instagram (HikerAPI), plus the existing Google Maps and vertical scrapers. A Sonnet-based source router picks which sources to invoke per campaign based on your ICP — no manual configuration. Routing accuracy hit 100% on the launch eval across five verticals (local physical, hospitality, visual brand, online B2B, long-tail).
- March 2026
Enrichment waterfall + cross-tenant cooldown
Two-tier enrichment kicks in for qualified leads missing email: Findymail fills the gap at $0.05/hit (success-billed), then MillionVerifier validates the result. Plus cross-tenant cooldown: when any Overwise customer reaches out to a lead, that lead is hidden from all other customers' campaigns for 7 days (30 days if a reply comes back). No more two-customers-pitching-the-same-prospect overlap.
- March 2026
Brand-voice extraction from your sent folder
On mailbox connect, Overwise reads the most recent ~200 sent messages, picks 5 representative samples, and stores a per-project brand-voice fingerprint. Every future draft is scored against the fingerprint; drafts below 0.7 voice-match are regenerated. Settings → Brand Voice page lets you swap samples or edit the extracted persona.
- February 2026
Cross-tenant Public-Lead pool
Lead-data architecture migrated to a shared PublicLead pool with per-project overlays — every time one customer's discovery surfaces a domain, every other customer's enrichment, cooldown, and embedding work compounds on it. Practical wins: faster discovery, less duplicate scraping cost, and the foundation for cooldown + vector search that shipped in March and May.
Your first leads, in 5 minutes.
14-day trial. Card on file, hard-stop at your lead cap (no surprise bills). All trust artifacts standard. Cancel any time without an email.
Start 14-day trial — first leads in 5 min → Card on file · No demo gate · Cancel any time