Customer Insights — Every customer signal, one clear answer

Open Customer Insights

Search calls, support tickets, Slack conversations, and company context from one private intelligence layer.

Customer Insights is an open-source, Clerk-authenticated workspace with a web chat, company and competitor views, semantic search powered by Together AI, scheduled Convex ingestion, and an MCP endpoint for compatible agents.

This repository contains application code and synthetic demo data only. It does not include a hosted deployment, customer data, or credentials.

What you get

Grounded chat across call transcripts, support tickets, and Slack
Semantic and keyword search with source inspection
Company timelines and customer-status views
Competitor mention tracking
An authenticated MCP server for coding agents and assistants
Optional Gong, Pylon, Slack, and company-enrichment integrations
A deterministic local seed dataset using reserved .example domains

Architecture

Gong / Pylon / Slack
        |
        | optional Convex ingestion jobs
        v
 conversations + text chunks + company context
        |
        | Together AI embeddings
        v
 Convex hybrid search
        |
        +--> authenticated web chat
        +--> company and competitor views
        +--> authenticated MCP endpoint

Each external integration is optional. Jobs without matching credentials skip safely, so you can begin with synthetic data and add only the sources you use.

Run locally

Prerequisites

Bun
A Convex project
A Clerk application
A Together AI API key for chat and embeddings

1. Install and configure

bun install
cp .env.example .env.local

In Clerk, create a JWT template named convex. Add the Clerk keys, Together AI key, and a random INTERNAL_CONVEX_SECRET to .env.local.

2. Start Convex

In the first terminal:

bun run convex

Follow the Convex setup prompts on the first run. The command keeps the local backend running and updates generated types as Convex files change.

For a one-shot schema/function sync instead:

bun run convex:once

3. Load synthetic demo data

With Convex running:

bun run seed

The seed is deterministic and idempotent. It creates 12 fictional companies, 36 calls, 48 tickets, 84 searchable chunks, 24 Slack mentions, 24 timeline events, and a multi-vendor competitor leaderboard. It refuses to run over non-demo data unless you explicitly pass {"force":true} to the Convex function.

Remove only the synthetic rows with:

bun run seed:clear

4. Start the web app

In a second terminal:

bun run dev

Open http://localhost:3030.

Configuration

Core variables:

Variable	Purpose
`NEXT_PUBLIC_APP_URL`	Public app origin; keep `http://localhost:3030` locally
`NEXT_PUBLIC_CONVEX_URL`	Convex deployment URL
`NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY`	Clerk browser key
`CLERK_SECRET_KEY`	Clerk server key
`CLERK_JWT_ISSUER_DOMAIN`	Issuer for the Clerk JWT template named `convex`
`INTERNAL_CONVEX_SECRET`	High-entropy secret for trusted server-to-Convex calls
`TOGETHER_API_KEY`	Together AI key for chat, embeddings, and classification

Optional integrations:

Variable	Enables
`GONG_ACCESS_KEY`, `GONG_ACCESS_KEY_SECRET`	Gong call ingestion
`PYLON_API_KEY`	Pylon support-ticket ingestion
`SLACK_MCP_XOXB_TOKEN`	Slack search and enrichment
`ORGANIZATION_EMAIL_DOMAINS`	Domains used to identify internal participants
`EXA_API_KEY`	Company enrichment
`SALES_WINS_SLACK_CHANNEL_ID`, `SALES_WINS_INITIAL_TIMESTAMP`	Optional closed-won Slack ingestion

See .env.example for the full template.

Load real data

Scheduled jobs live in convex/crons.ts. Historical imports and embedding helpers live in scripts/:

bun run scripts/backfill.ts gong 365
bun run scripts/embed.ts

Imports can contain sensitive customer data. Local databases, exports, transcripts, and tickets are ignored by Git and must never be committed.

MCP server

The authenticated MCP endpoint is:

https://your-deployment.example/api/mcp

It uses Clerk OAuth and exposes tools for hybrid search, full source retrieval, company discovery, and configured Slack context. The signed-in web app shows connection details for supported clients.

Commands

bun run dev          # Next.js on port 3030
bun run convex       # Keep the Convex dev backend running
bun run convex:once  # Sync Convex once and exit
bun run seed         # Add synthetic demo data
bun run seed:clear   # Remove synthetic demo data
bun run test         # Unit tests
bun run typecheck    # TypeScript checks
bun run build        # Production build

The cover image is reproducible from public/cover/preview.html.