csr1010 / wappy-kit

# Wappy
### Build a real AI agent on WhatsApp, running on your own machine, in under a minute.
Free. Open source. No account with us, ever. Your conversations, your memory, your data stay in
one file on your own machine — nothing routes through a server we run.
## Why now
AI models finally got good enough to reliably use tools, remember context, and hold a real
conversation, not just autocomplete a reply. At the same time, WhatsApp is already the app on
over 2 billion phones, nobody has to install anything new to talk to your agent. Put those two
things together and the obvious move is: skip building a new app, skip handing your data to
someone else's cloud, and just run your own agent, locally, on the channel people already have
open.
That's what this is. One command, your own model key, your own WhatsApp number, your own data.
## Install — under a minute, free, 100% local
 [code] 
 [code] 
A few questions, then a working agent. No credit card, no signup for Wappy itself, nothing
phoning home.
## Personal use cases
- 🧠 **A personal assistant that actually remembers you.** Not just the last message, a real
 session profile of what's going on, so three days later it still knows what you meant.
- 📅 **"What's on my calendar / find that email"** — connect your own Google account once, then
 just ask in plain English. No fixed command list.
- 📚 **A private, grounded knowledge base.** Ingest your own notes or documents and ask questions
 about them, answered from what you actually gave it, never a hallucinated guess, and it never
 leaves your machine.
## Business use cases
- 🎫 **A support agent** that resolves things in your own ticketing system instead of collecting
 an email and vanishing.
- 🍽️ **A booking agent** for a restaurant, clinic, or studio, wired to your real reservations
 system.
- 🏢 **An internal ops bot** for IT or HR, reachable from the app your team already has open all
 day.
None of these ship pre-built, they're what you wire your own data and tools into. The agent
(memory, context, tool-calling, reply formatting) is already built; your business logic is yours.
## Memory backend: local or Cognee
Every generated project wires in a real `Knowledge`/RAG layer, so anything you ingest can
actually be retrieved and grounded in a reply. The interview picks what backs it:
- **Local (default)** — SQLite/LibSQL, BM25 matching, zero setup, fully offline.
- **Cognee** — self-hosted or cloud. Extracts entities and relationships
 into a real knowledge graph instead of plain keyword matching, and keeps refining it over time.
 Needs your own running Cognee instance — the generated project's `COGNEE_SETUP.md` covers both
 the self-hosted and cloud paths.
Same rule as everywhere else here: bring your own instance, nothing routed through us.
## A real sample tool: Gmail/Calendar assistant
Say "yes" to the productivity question and you get a working example of the tool-calling harness
used for everything else here, not a toy demo. Connect your own Google