Simon-Pierre Boucher
All Groupe Ka projects
ka2 icon
Groupe Ka
AI Explorer Bots

ka2

The light edition of the Quebec web knowledge-graph explorer

An AI-driven scraper + search bot mapping the Quebec web as a knowledge graph — entities, attributes and typed relations — at the economical tier: Haiku-controlled, 6 concurrent, ~800 pages/day.

Haiku 4.5

Controller model

6

Concurrent missions

~800

Pages / day

Lime

Theme

Overview

ka2 is an AI-driven scraper and search bot that maps the Quebec web as a knowledge graph: entities (companies, organizations, people, creators, agencies, brands…), their attributes and their typed relations. Groupe Ka operates three twin bots sharing the same engine at three tiers — ka2 is the light, economical edition (lime-green theme), controlled by Claude Haiku 4.5 with 6 concurrent missions and about 800 pages a day.

Missions launched from the live dashboard drive a Claude agent that plans, searches and deep-dives through tools — search_web, map_site, scrape_page, deep_dive, crawl_queue. Fetching goes through Firecrawl with a Scrapfly anti-bot fallback, guarded by robots.txt, rate limits, fail-fast and a circuit breaker. A Haiku extractor turns each page's markdown into graph fragments — entities and typed relations — which pass through entity resolution with provenance before landing in SQLite (WAL).

The result is browsable on the public dashboard at ka2.bot: a real-time board and a structured directory with an interactive global graph.

Key Features

Agent-driven missions

A Claude agent plans, searches and deep-dives autonomously, steering the crawl through a tool loop instead of fixed scripts.

Knowledge-graph extraction

A Haiku extractor converts scraped markdown into entities and typed relations, with entity resolution and per-fact provenance.

Resilient fetching

Firecrawl primary with Scrapfly anti-bot fallback, robots.txt respect, rate limiting, fail-fast and a circuit breaker.

Live dashboard

Real-time board plus a structured directory with an interactive global graph, public at ka2.bot.

Tech Stack

Engine

Python
FastAPI
Claude (agent + extractor)
SQLite (WAL)

Fetching

Firecrawl
Scrapfly fallback
robots.txt + circuit breaker

Platform

ngrok → www.ka2.bot
shared engine with ka4/ka6

ka2 is live

Quebec web explorer — light edition — part of the Groupe Ka ecosystem.

www.ka2.bot