A team messenger
with AI as first-class members.
Channels for how your team already works. Iris, Ori, Kai, and Sage show up as members — reading, remembering, contributing when useful, staying quiet otherwise. Everything else you run plugs in over the open Choir Agent Protocol.
No credit card. Free during open beta.
Anyone know why the ETL job is behind on prod?
The overnight run hit a rate limit at 03:41 UTC — retry scheduled at 08:00. Grafana panel: api-throttling.
grounded in runbook.on-call · docs.eng.observability
Thanks — @ori remind me in an hour to check.
Reminder set: 'check ETL retry' at 15:47.
Everything you need on day one
Not a starter kit. The actual thing.
Real channels, familiar shape
If you've used Slack you already know the surface. Public and private channels, DMs, threads, mentions.
- Web + iOS/Android + desktop
- Rich text, attachments, reactions, pins
- Search across every message + attachment OCR
Four AI members on day one
Iris answers product/ops questions. Ori sets and fires reminders. Kai remembers what matters. Sage briefs you each morning.
- Each has a persona, tool kit, and memory you control
- Add them to channels like any teammate
- Off / quiet / active per Voice, per channel
Grounded on your knowledge
Voices read your workspace KB before answering. When the grounding isn't there, they say so — never invent.
- Write articles in a plain editor
- Every miss becomes a gap in the admin dashboard
- Attachments auto-OCR into search
Reminders + cadences
"@ori remind me tomorrow to review the migration" — Ori fires it into DM at the right time. Cadences repeat weekly / monthly / on-call.
- Per-user notification prefs
- Push notifications on mobile
- Snooze / repeat / hand off
CAP integrations
Your CI, your data warehouse, your on-call system — anything that speaks CAP posts and reads like a Voice. Open protocol, two-language SDK.
- Signed envelopes (ES256 detached JWS)
- Channel-scoped permissions + audit log
- Downbeat and Zap Bridge as first-party partners
Roles that fit real orgs
admin / manager / member / guest / driver — plus per-workspace identity (title, bio, expertise) so Voices address people by what they actually do.
- Invitations by email + link
- Per-workspace SSO (Google, Apple)
- Staff-tier admin for Choir operations
@iris what's our current WhatsApp deliverability policy for order confirmations?
Order confirmations go through the transactional utility template order_confirm_v2 — free-tier conversation, no session opt-in needed. Marketing messages need explicit opt-in per Meta's 2024 policy update.
The grounded difference
Iris cites her sources. Every reply.
When a Voice can answer from your KB, she does — and shows exactly which articles she read. When she can't, she says so and surfaces the gap to your admin dashboard so you can fill it.
- Every KB source appears inline with the reply
- Missing-knowledge gaps become one-click "write this article"
- Zero hallucinated policy, code paths, or numbers
How you get there
Set it up in minutes, not weekends.
- 01
Sign up + say "team"
One sentence to the onboarding wizard, or pick the Teams card.
- 02
Land in #general
Iris + Ori + Kai are already there. Chat like you would anywhere else.
- 03
Invite your team
Email invites, or share the join link. Everyone gets an identity gate on first entry so Voices know who's who.
- 04
Plug in the rest
CAP for anything with an API. Or start with the built-in tools — reminders, briefs, memory — and integrate later.
Answers
What people ask us most.
How is this different from Slack + a ChatGPT tab?+
The AI is inside the channels, grounded in your data, with a memory that persists across turns. It doesn't forget between chats. It doesn't hallucinate about your codebase or your KB — it says "I don't know" and surfaces the gap.
Can we run our own tools inside Choir?+
Yes — the Choir Agent Protocol (CAP) is open. Your tools post + read as first-class members. Two-language SDK (Python + TypeScript). Signed envelopes, channel-scoped grants, full audit trail.
What about pricing?+
Free during open beta. Post-beta pricing will be seat-based with a free tier for small teams.
Is my data used to train models?+
No. Grounding happens against your KB at request time; nothing is fine-tuned on your data.
Ready to try it with your team?
Two minutes to sign up. Ten minutes to have a real conversation with Iris. Invite the rest of the team when you're ready.
Free during open beta.