Chord is a multiplayer AI workspace for sales teams. The unit is a room — one per account or open deal. Everyone working that account (AE, SDR, SE, sales manager) lives in the room. Five AI agents live in the room too: @research, @objection, @coach, @drafter, @deal-pulse. Everyone reads everyone else's context in real-time.
The thesis: 1-human-1-AI tools (ChatGPT, Cursor for sales) collapse when a deal has six humans pulling on it. Knowledge gets siloed in private chats, prep notes go stale, and the AI never builds a memory of the account. Chord makes the AI a shared resource — when @research finds out the buyer's CFO just left, the entire deal team sees it the same instant.
Architecturally it's a real-time collaboration substrate (think Liveblocks-style CRDTs) layered over a Salesforce/HubSpot sync. The AI agents are Claude-backed, scoped to a single room's context window, and they're triggered by @-mentions, file uploads, or scheduled probes (e.g. @deal-pulse runs a stalled-deal scan every morning). Pricing is $40/seat/month plus a 1% success fee on closed-won deals — aligned with revenue, not seats.
The interesting product question is permissioning — sales managers see all rooms, AEs see only their book, and the AI agents inherit the requesting user's CRM permissions on every call. We model this as Postgres RLS on the room table, with the AI agents acting as a 'service user' that proxies the requester's identity.
- 01Rooms per account with persistent shared context across team + AI
- 02Five specialized agents — @research, @objection, @coach, @drafter, @deal-pulse
- 03Live CRDT-backed multiplayer surface over Salesforce / HubSpot sync
- 04Postgres RLS — agents inherit requester's CRM permissions
- 05$40/seat + 1% success fee — outcome-aligned pricing
More work
Seven other projects across crypto, AI, and games.