Simon-Pierre Boucher
All Groupe Ka projects
API·Ka icon
Groupe Ka
Data & Infrastructure

API·Ka

The data layer of the ecosystem — collectors, API, and the KA Agent

Nine daily collectors persist every platform's data into unified PostgreSQL, exposed through a public FastAPI — plus the ecosystem's conversational agent and connector supervision.

9

Daily collectors

63/63

Collections OK (7 days)

4.16M

Records / 7 days

~3,270

Connectors supervised

24+

KA Agent tools

Overview

API·Ka is the central data platform of the Groupe Ka ecosystem. Every day, nine collectors (Lou·Ka, Immo·Ka, Food·Ka, Auto·Ka, Fabri·Ka, Resto·Ka, Sorti·Ka, Créa·Ka, Job·Ka) persist the platforms' data into a unified PostgreSQL base — fetch → validation → checksum → insert → backup → run journal, with 3× retry and exponential backoff — then expose it through a public FastAPI: paginated data, per-date history, statistics and PDF reports.

It also hosts the KA Agent — the group's central AI assistant (Claude Haiku 4.5, SSE streaming, a tool loop of 24+ tools wired to the platforms' live public APIs) served as an embeddable widget to the ecosystem's 13 domains — and acts as the connector supervisor: roughly 3,270 connectors tracked with collection history, per-service health and alerts.

Over a recent 7-day window: 63 of 63 collections succeeded, 4.16M records collected, 6,196 API calls served. Product API access requires authentication — a KA ID session or a personal kapi_ bearer token generated on groupe-ka.com — while monitoring, stats and the agent remain open. A daily backfill catches missed dates, and timestamped per-service backups are retained for 90 days.

Key Features

Daily collection pipeline

Nine parallel independent collectors at 02:00 with validation, checksums, 3× retry with backoff, run journal and automatic 7-day backfill.

Public data API

Paginated per-service data, latest and per-date snapshots, stats and run history — secured by KA ID session or kapi_ bearer token.

KA Agent

Claude-powered SSE chat with a growing tool loop (24 → 33 tools) over live platform data, served as an embeddable widget to all 13 domains.

Connector supervision

~3,270 connectors across the ecosystem tracked with per-service health, statuses (ok/degraded/broken/retired) and alerts.

Ecosystem PDF reports

Per-platform dashboards and a consolidated ecosystem report, plus custom report building from a block catalogue.

Native-app auth exchange

SSO endpoints for KA ID plus a token exchange for the iOS/Android apps (HS256, audience-scoped).

Tech Stack

Backend

Python 3.11+
FastAPI + Uvicorn
PostgreSQL
SQLAlchemy 2 + Alembic
APScheduler

AI

Anthropic SDK
Claude Haiku 4.5
SSE streaming
fuzzy search cascade

Platform

PM2 (api / scheduler / ngrok)
daily backups (90-day retention)
pytest

API·Ka is live

The ecosystem's data, by API — part of the Groupe Ka ecosystem.

www.api-ka.com