
Competitive Intelligence
v0.2.0A competitor watch that starts empty and asks you who to track. Competitor Scout captures each rival's site, pricing, and changelog as dated evidence (text and screenshots), turns the diff into a sourced signal log, and ships a weekly PDF briefing you can forward or have emailed to you. Deterministic scripts do the fetching, diffing, screenshotting, and rendering; the agent only spends its judgment on what a change means.
Competitive Intelligence
agents
Competitor Scout
.agents/competitor-scout/
Daily Sweep
0 7 * * 1-5
Monthly Landscape Review
0 9 1 * *
Weekly Report
0 8 * * 1
1
Agents
3
Jobs
1
Depts
2
Pages
Runs the competitor watch end to end — scripted daily capture, sourced signal log, and a weekly PDF briefing with the "so what for us" on every material move. Starts every cabinet empty and asks who to track before doing anything else.
Competitive Intelligence
An always-on competitor watch that starts completely empty. One agent, Competitor Scout, asks who to track, then a set of scripts does the daily capture (text and screenshots) for free, and the agent spends its judgment on one thing: turning a week of sourced evidence into a weekly PDF briefing you can open, forward, or have emailed to you automatically.
Keep your CRM, your BI, your Slack. Replace the competitor spreadsheet nobody updates, the battlecards that go stale, and the "what did they just ship?" fire drill before every deal.
Why this cabinet is built the way it is
Most competitive-intel tools either cost a fortune in API calls (an LLM re-reading full competitor pages every day) or ship full of fake demo data that has to be torn out before it's useful. This cabinet does neither:
- It ships empty. No placeholder competitors, no fabricated pricing tables. The first thing Competitor Scout does is ask who you actually want tracked.
- Plain automation does the expensive part for free. Fetching pages, diffing them against yesterday, taking screenshots, and rendering the HTML/PDF report cost zero LLM tokens. The agent only reads a compact summary of what changed and writes the one thing automation can't: what it means for you.
- No social-media API required. Every source is a public page — homepage, pricing, changelog, blog, careers. Screenshots come from a locally installed Chrome running headless, not a paid scraping service.
- The sidebar stays out of your way. Everything this cabinet needs to run day to day is tucked out of sight. What you see is [[competitors/index]], [[dashboard]], and [[briefing]] — the things you'd actually want to open.
The team
- [[.agents/competitor-scout]] — Competitor Scout. Onboards you, runs the scripted daily sweep, decides what's material, and ships the weekly briefing. One agent, because most of the work isn't agent work at all.
Recurring rhythm
| Cadence | Job | What happens |
|---|---|---|
| Weekdays 07:00 | Daily sweep | Fetch, diff, screenshot every due competitor. Pure automation; asks for onboarding details instead if nothing's tracked yet. |
| Monday 08:00 | Weekly report | Agent reads the week's evidence, decides what's material, writes the briefing, renders the PDF, emails it if a recipient is set. |
| 1st of month, 09:00 | Monthly landscape | Compiles the month's weekly briefings into a board-grade landscape review. Cheap: it reads already-synthesized material, not raw pages. |
Get started
- Open the [[dashboard]] — empty right now. It fills in after the first sweep.
- Tell Competitor Scout who to track. Message it in chat with your company and a few competitor names.
- Optionally tell it who should get the weekly PDF by email. Without a recipient, the report still gets written; you just open or forward it yourself.
- From there it runs itself. Check back on the [[dashboard]] for the glanceable view, or open [[briefing]] for the latest report.
Connectors
Optional: Gmail (better-with, send only) — emails the weekly PDF to whoever's
configured. Everything else in this cabinet works with zero connectors: public web pages
and a local Chrome install are all the capture step needs.
$ git clone --filter=blob:none --sparse https://github.com/cabinetai/cabinets.git && cd cabinets && git sparse-checkout set competitive-intelligence