Launching today

CustomerDog — Customer Support Agent
Clone, deploy, configure your knowledge base, embed in 5min
3 followers
Clone, deploy, configure your knowledge base, embed in 5min
3 followers
Open-source agentic customer support. Clone, deploy, configure your knowledge base, embed on your site. CustomerDog answers visitor questions instantly from your own docs, files tickets, sends follow-up emails, and looks up customer data from Stripe, Shopify, Linear, and 100+ apps. Configure in minutes — drop your website URL, docs URL, and custom answers (refund policy, FAQs). It crawls your pages, blogs, and docs into a smart knowledge base the AI answers from.









I wanted an AI customer support agent for my own product, but every SaaS option started at $200+/month and locked the knowledge base, branding, and conversation history into their dashboard. So I built one as a clonable Next.js app: 7 env vars + Vercel deploy + paste your docs URL + done. MIT licensed.
What it ships:
Streaming chat (SSE) at /chat. Embed it on any site as a <script> bubble (/widget.js), an iframe (/embed), or just link to /chat.
Anonymous visitor sessions via cookies — no signup, no auth flow. Conversations correlated via qlaud's end_user_id.
Knowledge base — paste a docs URL (server fetches it and Mozilla Readability extracts the article body), paste markdown, or crawl a whole sitemap with one click. Optional Firecrawl integration for JS-rendered SPAs.
Admin UI for KB management, past-conversation transcripts, embed snippet generator. Single-password gate (no Clerk/Auth0).
Architecture is deliberately small (~3,500 LOC, 3 Supabase tables). Tools live entirely at qlaud (an Anthropic billing/runtime layer): built-ins for Resend, Slack, Linear, Zendesk, GitHub, Notion, Twilio, plus an MCP catalog with Stripe, Shopify, HubSpot, Cal.com, etc. customerdog sends tools_mode: "tenant" and qlaud auto-attaches whatever the operator tenant-shared in their dashboard. Streaming + tool dispatch share a single SSE — Anthropic's standard content_block_* events interleaved with qlaud's tool_dispatch_start/done events.
What it deliberately isn't: a multi-tenant SaaS, an agent framework, a vector DB. The KB lives as plain text concatenated into acache_control: ephemeral system prompt — Anthropic's prompt cache makes the long context cheap (~10% of an uncached turn), so no embedding pipeline, no retrieval. Works great for a typical B2B SaaS docs corpus (~20-80K tokens). Bigger KBs would want retrieval, which is straightforward to bolt on.
Setup is genuinely 5 minutes:
Click the Vercel deploy button → fill in 7 env vars (qlaud key, three Supabase values, two random secrets, your deploy URL).
First admin page load auto-runs schema.sql against Supabase via direct Postgres.
Paste your docs URL in /admin/kb (or crawl a whole sitemap).
In your qlaud dashboard, enable a couple of built-ins (e.g. Resend for email, Linear for tickets) and tenant-share them.
Drop the <script> snippet on your site.
Code: https://github.com/customerdog/customerdog
SIte: https://customerdog.com
try live at https://www.customerdog.com/chat built from https://qlaud.ai customer support