Guided tour
See 4399 CRM in three short walkthroughs
Screen recordings of the live site (2:44 in total, no sound, captions on screen): adding people, logging a meeting on the map, and the optional AI assistant with a person deciding what is kept. Below them, screenshots of every key feature. Prefer to try it yourself? A guest sandbox takes one click.
Workflow 1
Contacts
Open a private guest sandbox with one click, search the address book, and add people by user name and by scanning a QR code.
0:42 · no sound · captions are on screen MP4, 819 KB Captions (WebVTT)
Steps and transcript
Workflow 2
Log a meeting
Record who you met, when and where: search for the venue, check the pin on the map, then find the meeting on the records map and the calendar.
0:55 · no sound · captions are on screen MP4, 1.8 MB Captions (WebVTT)
Steps and transcript
Workflow 3
AI with a human in the loop
The optional meeting-note assistant with your own key: see exactly what would be sent, review an AI-labelled draft, accept it, then check the AI log, the activity log and your data export.
1:07 · no sound · captions are on screen MP4, 1.6 MB Captions (WebVTT)
Steps and transcript
Screenshots
Every key feature
Desktop at 1440 × 900 (light and dark), phones at 390 × 844. Select a screenshot to see it full size; the arrow keys move between them.
How these were made
- A Playwright script drives Google Chrome through each journey on the live site (last recorded on 6 October 2026). The same script is an end-to-end test: it fails if a step does not work. Captions and the cursor highlight are added by the script; long network waits are trimmed.
- The workflows run in a fresh guest sandbox, so the people and meetings are the generated sample data; Lena Park and the meeting with Ava Chen are added during the recording. The screenshots use the shared demo account and change nothing.
- The AI response in workflow 3 is mocked for illustration. The request to the provider is intercepted inside the test browser and answered with a fixed reply that says it is mocked, and the “key” typed is a placeholder, so no model was called and no key was used. The redaction preview, the labels, the decision buttons and the AI log are the real app.
- Reproduce it with
pnpm showcaseinweb/(see the README).