Guide

From first launch to a digest in your inbox. The checklist updates itself as you go.

7 / 7 complete

Starting the app

Double-click start.bat in the project folder. It starts the server and opens this page. Keep that window open while you work; close it to stop the app.

Manual alternative: python -m uvicorn app.main:app --port 8000, then open http://127.0.0.1:8000.

Step 1

Check your keys

Overview →
Keys live in the .env file (never in the app). The checklist on the right shows whether they are all in place. After changing .env, restart the app.
Step 2

Approve the company profile

Company profile →
The profile is the only source of JADA facts. Ideas and relevance scores are built from it, and the agent never invents positioning. Review the YAML, correct anything, set status: approved and press Validate & save.
Step 3

Discover competitors

Competitors →
Press ✨ Run discovery (about a minute). It searches the web from your profile and proposes ranked candidates. Each card shows an overlap score, a category (direct / indirect / adjacent), a rationale, and source links for the evidence. Candidates without a real source link are discarded automatically. You can also + Add manually.
Step 4

Approve, reject or park (the human gate)

Competitors →
Only approved competitors are ever monitored. ✓ Approve adds them and auto-detects their sources (blog feed, key pages, news). Watch later and Reject are remembered. Start with 3 to 5 competitors so the first digest is easy to judge.
Step 5

Confirm each LinkedIn page

Approved tab →
Open the Approved tab. If a competitor shows an amber Confirm banner, click the link, check it is really their company page, then press Confirm. LinkedIn posts are pulled through Apify only after you confirm, so the wrong page is never monitored. You can add, disable or enable sources per competitor here.
Step 6

Run the first collection and digest

Overview →
On Overview → Run a digest, choose Daily, tick First-run preview (last 7 days), leave Send to recipients unticked and press ▶ Run now. The stages stream live: collect → analyse → synthesise → ideas → QA → deliver. The first run can take several minutes because it also collects the 90-day baseline.
Why the preview tick? A competitor's first collection is a baseline: old posts are stored as context but never reported as "new". The tick lets you see the last 7 days once. From the second run on, digests only show posts published since the previous run.
Step 7

Review the digest before anyone gets it

Digests →
Open the digest. You see the email as recipients will (toggle Desktop / Mobile, or Plain text). On the right:
  • QA gate: citation, quote, number, labelling, JADA-fact, compliance and render checks. Anything that fails is removed, not sent.
  • Evidence lineage: every claim maps to stored items (see them in the Activity feed).
Things to look for: every statement links to a source, interpretations are labelled Our read, and the Data notes block lists anything unavailable.
Step 8

Add LinkedIn posts by hand (optional)

Manual input →
See a post worth including? Paste its URL and text under Manual input. It is deduplicated, analysed and used in the next digest. Pasted text is treated as data only.
Step 9

Send it

Digests →
In the digest's Delivery panel, type a test address and press Send this digest. Sending goes through Brevo. Once sent, the same posts are never reported again and a second send is refused unless you press Resend (force). To send straight from a run, tick Send to recipients on Overview.
Step 10

Day-to-day use

Runs & audit →
  • Daily / weekly: pick the cadence on Overview and press Run now. (Automatic scheduling is not built yet.)
  • Feedback: the 👍/👎 links in each email teach the topic weighting.
  • Something failed? Runs & audit shows every stage, cost and error. A source that fails 3 times in a row is switched to failing and skipped; re-enable it from the Competitors page.
  • Monthly: run discovery again. It only proposes new names and never removes approved ones.
  • Remove a competitor: Delete all data on its card erases it and everything collected about it.

What the labels mean

★ notable launch, pricing change, new market, partnership, funding…
Our read an interpretation, always with its evidence links
baseline older context, not reported as new
failing a source that failed 3 times; skipped until re-enabled
search-backed a candidate whose evidence came from a real search result