Real-time AI coaching for B2B sales teams.
- Live AI co-pilot during every B2B sales call — surfaces context from CRM/email/past calls, objection handlers calibrated to the prospect, and the next best move from your VP Sales's playbook.
- Invisible to screen share — Electron
BrowserWindow.setContentProtection(true)wrapping macOSNSWindowSharingNone+ WindowsWDA_EXCLUDEFROMCAPTURE. Tested across Zoom, Google Meet, Microsoft Teams, Loom, OBS. - Sub-2s p95 latency end-to-end — Recall.ai → Gladia ASR (FR primary, Recall built-in fallback) → Anthropic Haiku 4.5 streaming.
- Closed feedback loop adapted from Karpathy's
autoresearch— every Monday the VP Sales reviews 3–5 key call moments, each correction becomes a structured rule with a strict lifecycle (draft → published → observed → promoted | narrowed | rewritten | archived), rules auto-apply to every rep's next call. AI proposes, HoS disposes.
Stack: Next.js · TypeScript · Postgres + pgvector · Vercel AI SDK · Anthropic · Composio · Recall.ai · Gladia · Electron (desktop overlay) · Turborepo
Geography: Paris → San Francisco mid-2026. EU-native (Frankfurt data residency).
Team: Gauthier Girault (engineering, solo end-to-end), Clément Milard (sales).
Source: private. DM @gauthiergirlt for access.