Quorum

Meeting intelligence for revenue, customer success and engineering teams. Quorum turns recorded calls into notes you can trust: an executive summary, the decisions that were made, the risks that were raised, who owes what by when, and a transcript you can search and jump around in. Ask a question about one meeting or your whole library and get an answer that links to the exact moment it was said.

Quorum is a rebuild of the core of Fathom, scoped to the parts that matter after the call ends.

Live: https://quorum-production-91b5.up.railway.app

What you can do

Home	Library stats, the next meetings on your calendar with auto-record toggles, recent meetings, action items due soon and where meeting time goes.
Upcoming	Your calendar for the next two weeks. Connect Google Calendar (read-only) or use the demo calendar; set meeting preferences (auto-record external / internal) and toggle recording per meeting.
Action items	Every follow-up across meetings, filterable by status and owner, with overdue items flagged.
Ask Quorum panel	Always available on the right (desktop), like Ask Fathom: ask across all meetings or a scope, and the conversation follows you between pages.
Meetings	Filter by title, company or participant; narrow by type (Sales, Customer Success, Internal, Engineering); sort by date or length.
Meeting detail	Header with participants and platform · conversation timeline with talk-time per speaker · summary with key points, decisions and concerns (each linked to its source line) · sortable, filterable action items you can tick off · timestamped transcript with in-transcript search (Ctrl/⌘+F), highlights and click-to-jump.
Ask AI (one meeting)	Chat in a side panel (bottom sheet on mobile). Answers come with cited transcript lines and a confidence level; clicking a source scrolls the transcript to that line.
Ask Quorum (all meetings)	Pattern questions across the library ("What issues come up repeatedly in sales calls?"), scoped to all meetings, sales, customer success or internal. Sources open the meeting at the cited timestamp.
Global search	Ctrl/⌘+K (or /) from anywhere: meetings by title, company, participant, and every transcript line that mentions your words, with the match highlighted. Enter opens the meeting at that moment. A full results page lives at /search?q=.
Regenerate summary	Re-runs the notes for a meeting through the model on demand.

The library ships with eight realistic meetings from a fictional company, Lattice Labs: four sales calls, two customer-success reviews, an architecture review, a roadmap planning session and an incident postmortem. The same customers and problems recur across them, so cross-meeting questions have real answers.

Stack

• Next.js 16 (App Router, React 19) + TypeScript
• Tailwind CSS v4, Motion for the few transitions that help (drawers, sheets, palette), Lucide icons
• Anthropic API via the official @anthropic-ai/sdk, called only from server route handlers
• No database: the seed library in lib/data stands in for one

Running locally

git clone https://github.com/wasaybuilds/quorum.git
cd quorum
npm install
cp .env.example .env.local   # then add your key
npm run dev                  # http://localhost:3000

ANTHROPIC_API_KEY is optional. Without it every AI feature still works using the offline fallback described below, and answers are labelled "Offline answer".

Variable	Required	Purpose
ANTHROPIC_API_KEY	No	Enables live summaries and answers. Server-side only.
ANTHROPIC_MODEL	No	Overrides the default model.
GOOGLE_CLIENT_ID / GOOGLE_CLIENT_SECRET	No	Enables "Connect Google Calendar" on the Upcoming page.
DATABASE_URL	For accounts	Postgres connection string. Enables sign-in and per-user saved data.
AUTH_SECRET	For accounts	Encryption key for Google tokens stored in the database.

Production build: npm run build && npm start (respects PORT).

How the AI works

Browser ──POST──▶ /api/ai/* (Next.js route handler) ──▶ Anthropic Messages API
                         │                                   │
                         │◀──── structured JSON (schema) ────┘
                         └── on any failure: grounded offline answer

The API key never leaves the server. All three routes use structured outputs (a Zod schema per route), so responses are always well-formed JSON.

Route	Input	Output
POST /api/ai/analyze-meeting	{ meetingId } or { transcript, title, participants }	executive_summary, key_points, decisions, concerns, action_items[] { task, owner, dueDate, priority }
POST /api/ai/ask-meeting	{ meetingId, question, history? }	answer, sources[] { timestamp, speaker, text }, confidence
POST /api/ai/ask-multi-meeting	{ question, meetingIds?, history? }	answer, patterns[], sources[] { meetingId, meetingTitle, timestamp, text }

Citations are line numbers, not quotes. The transcript is sent with a line label on every entry ([L12] 4:31 Sarah Chen: …) and the model returns the line numbers that support its answer. The server maps those back to real transcript entries, so a cited source can't be a paraphrase or a hallucinated quote, and every source is clickable.

Offline fallback. If there's no key, the request fails, or the response is unusable, the routes answer from the meeting's own data instead of showing an error. The fallback works out the intent (concerns, decisions, next steps, overview) and ranks transcript lines by keyword to find supporting sources; for cross-meeting questions it counts recurring themes (security review, Salesforce sync, pricing, timeline, sign-off) across the scoped meetings. The UI marks these answers so it's clear which is which.

Project layout

app/
  page.tsx                  dashboard
  meetings/page.tsx         meetings list
  meetings/[id]/page.tsx    meeting detail (statically generated per meeting)
  ask/page.tsx              Ask Quorum across meetings
  search/page.tsx           full search results
  api/ai/*/route.ts         analyze-meeting, ask-meeting, ask-multi-meeting
components/
  layout/                   sidebar, mobile nav, page header
  dashboard/                stats, recent meetings, side panels
  meetings/                 list + card
  meeting-detail/           header, timeline, summary, action items, transcript
  ask/                      chat thread, answer renderer, source cards
  search/                   ⌘K palette, results page, highlighting
  common/                   avatar, badges, loading / empty / error states
lib/
  data/                     seed meetings + calendar stub
  ai/                       model client, prompts & schemas, offline fallback
  utils/                    formatting, search & ranking

What's real and what's stubbed

Built: everything in "What you can do" above, responsive from 375px phones to wide desktops, with loading skeletons, empty states and error states with retry.

Deliberately out of scope:

• Recording and transcription. No bot joins calls; meetings come from the seed library.
• Video playback. The timeline and timestamps navigate the transcript rather than a video.
• Calendar-driven recording. Your real Google Calendar is read and record preferences are saved, but no bot joins the meetings.
• Shared workspaces. Every account sees the same seed meeting library; teams, sharing and per-user recordings are not modelled.
• CRM / Slack / Asana sync, clip sharing, custom summary templates, coaching metrics.

Accounts and data

Sign-in is Google OAuth (OpenID Connect), done in-house in lib/calendar/google.ts and app/api/auth/*. One consent screen both identifies the user and grants read-only calendar access.

• Sessions: a random token in an httpOnly cookie; the database stores only its SHA-256 hash (30-day expiry). Sign-out deletes the row.
• Google tokens: encrypted with AES-256-GCM (AUTH_SECRET) before they're written; refreshed automatically; Disconnect deletes them and revokes the grant at Google.
• Per-user data (Postgres): users, sessions, Google tokens, recording preferences and per-meeting record toggles, and action-item completion. Tables are created automatically on first use (lib/db/index.ts).
• Demo workspace: "Explore the demo workspace" on the login page skips sign-in for evaluators; changes then last only for the visit.
• Gate: proxy.ts sends visitors without a session or demo cookie to /login; every API route re-checks the session server-side.

Google Calendar setup (optional)

1. In Google Cloud Console, enable the Google Calendar API and create an OAuth client ID of type Web application.
2. Add https://<your-domain>/api/calendar/callback (and http://localhost:3000/api/calendar/callback for local use) as an authorised redirect URI.
3. On the OAuth consent screen, add the calendar.readonly scope. While the app is in Testing, add each Google account that should connect as a test user; others will see Google's "unverified app" screen.
4. Set GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, DATABASE_URL and AUTH_SECRET, then restart.

The same OAuth client powers sign-in; the callback path /api/calendar/callback is the only redirect URI needed.

Deployment

Deployed on Railway as a standard Node service: build npm run build, start npm start. Set ANTHROPIC_API_KEY in the service variables.