All automations
Research & monitoring3–4 weeks to livefrom €5,250

Market & filings watcher

Most monitoring dies because a human has to remember to look. I run this one myself: it has been live since June 2026, its first autonomous run sent 17 alerts with 0 errors, and it writes state around 270 times a day. It looks every five minutes, stays silent until something clears a threshold, and never sends the same thing twice.

Live on this site – every enquiry runs through it
replay · illustrative timing · real pipeline
enquiry checked & cleaned0.4s
summarised into a brief1.8s
scored against your rules2.6s
logged, assigned, alert sent4.0s
a human writes the replyon purpose

How it works

Four steps, and only one of them is AI. Deterministic code decides what counts as a signal; the model is there to put it into words.

01 · fetch

Scheduled, and failure-tolerant

Scheduled functions pull prices, news feeds and public filings. A source going down degrades to empty – it never takes the watcher down with it.

02 · filter

Your rules, not a feed

Deterministic per-entity rules: include patterns, clickbait excludes, action triggers. Deduplication windows (24 hours for signals, 7 days for news) mean nothing arrives twice.

03 · summarize

The one AI step

A plain-language summary in your language. If the model is unavailable, the original headline goes out instead – the pipeline does not wait for the AI.

04 · deliver

Into your chat, and back

Alerts land in your chat. Commands come back through the same channel: status, pause, check this ticker. The rules live in git – every change to what the watcher does is a reviewable diff.

What you gain

  • You stop polling; the watcher polls
  • Every rule change is a reviewed commit, not a setting someone silently flipped
  • The AI step can fail without the watcher failing
  • New entities are validated against live data before they go on watch – not just typed into a box

Honest limits

  • Alerts are observations, not decisions – the human acts, and that is by design
  • It watches public and API-accessible sources; paywalled data is a separate conversation
  • State and code must live in separate repositories. We learned this the expensive way: our first version triggered a full rebuild on every state write – 840 hours of wasted build machine time in two months before we split them
Time to live
3–4 weeks
Price
from €5,250
Typical saving
approximately 5h/week
You own
code & data

indicative, not a quoted delivery date

What do you check every morning before anything else?

Tell me on a 30-minute call – if a rule can describe it, a watcher can carry it.

Book a discovery call