AI outreach assistant

AlatPintar

Finds local businesses on Google Maps, scores how well they fit what you sell, and drafts the first WhatsApp or email message in your style. A person approves every message before it is sent. Free to self-host (AGPL) or use the hosted version.

AlatPintar: prospect finder with AI-written first messages

The problem

Reaching local businesses means finding them, researching them and writing each first message yourself, then remembering to follow up.

What it does

  • Collects businesses from Google Maps with a headless browser (Playwright); other sources include OpenStreetMap, Overture and CSV or Excel import.
  • Scores each lead for fit and drafts a WhatsApp or email message with an AI model.
  • Every draft waits in an approval outbox; sending is paced to protect the phone number.
  • Learns the sender’s writing style from exported chats, keeping only summary figures.
  • Dashboard for campaigns, leads, outbox, pipeline, knowledge base and analytics, in English and Indonesian.
  • Free community edition (AGPL) to self-host; a hosted edition adds premium parts from a separate private package.

How it is built

  • NestJS API with Prisma and BullMQ job queues; Next.js web app; PostgreSQL with pgvector and Redis.
  • Turborepo monorepo, Docker images, nginx config and GitHub Actions for CI and image builds.
  • Playwright end-to-end specs plus unit tests inside the API.
  • Multi-workspace design: each business keeps its own prospects, messages and accounts.

How AI is used

  • Any OpenAI-compatible provider, chosen by environment (examples in the docs: OpenAI, OpenRouter, a local Ollama model).
  • Used for a business brief from a website, lead scoring, message and reply drafting, reply classification, writing-style profile and knowledge-base embeddings.

Honest notes

  • Sending through a linked-device WhatsApp library is not an official channel, so volumes should start small.
  • Nothing is ever sent without a person’s approval.

Engineering depth

Counts taken from the repository itself, not performance claims.

109
route decorators in 18 controllers
29
database models
23
migrations
247
GitHub stars, 69 forks