The browser that does the work.
Yours, end-to-end.
A Chromium browser with a native AI agent. Multilingual, on-device, MIT-licensed. No accounts required, no telemetry on by default, no prompts shipped to a vendor cloud.
Q3 enterprise AI adoption — research notes
Working memo across 6 vendor sites. The pattern that keeps surfacing: enterprises don't buy AI features anymore — they buy workflows. Pricing is shifting from seats to outcomes, and procurement teams are quietly aligning on a smaller, sharper bar for evaluation.
Three things I'd flag before the Friday review
- Per-seat pricing is being replaced with per-outcome / per-job credits. Build a "what does it cost to do one thing" view, not a seat count.
- Security posture matters more than benchmark performance for budget approvals. Procurement teams want SOC 2 + on-device inference options before a pilot.
- Two of the six vendors quietly shipped agent SDKs in the last 30 days. The vendor list is moving from "tools" to "platforms."
Vendor snapshot
| Vendor | Pricing | SOC 2 | On-device |
|---|---|---|---|
| Acme AI | $32 / seat / mo | Type II | partial |
| Northwind | $0.04 / job | Type II | yes |
| Helio | $28 / seat / mo | in audit | no |
| Praxis | $0.07 / job | Type II | yes |
| Quark | $45 / seat / mo | Type II | partial |
| Bulle | MIT · free | Type II | native |
Paris (CDG) → New York (JFK)
Showing fares for next Tuesday. Filtered to window seats, no red-eyes, return Friday. Three I'd call before booking:
Return: Fri JFK → CDG, evening departures only. I've pre-marked the AF return that pairs cleanly with this outbound. Say the word and I'll book + file the expense receipt.
Inbox — top of stack
I read all 47. Here are the seven that look like they actually need you. Drafts under two lines each — nothing sent without your nod.