The database is the source of truth for schedules
Queue state is derived and re-registered on every boot, so the job store can be flushed without losing a single watchlist.
How it works
A request goes in. Evidence comes out.
Every module rides the same five-stage pipeline. What changes per module is only the carrier handlers in the middle — which is why adding a carrier never touches the rest.
The pipeline
Ask. Orchestrate. Collect. Record. Notify.
- 01
Ask
Your TMS calls an authenticated route, or an operator fills a form. Instant requests run inline; scheduled ones register a Central-Time cron and are edited, paused or retraced later.
- 02
Orchestrate
Items are grouped by carrier and run concurrently through a handler registry. A carrier that is not wired returns a well-formed pending row instead of failing the batch.
- 03
Collect
A warm, self-healing pool of logged-in browsers and direct HTTP clients do the portal work — CAPTCHAs solved, sessions persisted, logins backed off when a portal is unhappy.
- 04
Record
Every run is written as request → execution → response with a portal capture as evidence, then normalized so a date means the same thing across every carrier.
- 05
Notify
A diff decides whether anything moved. Watchlists email only on change or as one daily digest, and ring the in-app bell; billing letters and release requests go out under your company identity with a Message-ID kept.
Design principles
The decisions that make it dependable.
Neo was built by running it against real portals for real loads. These are the rules that survived.
A warm, self-healing browser pool
Logged-in portal sessions stay alive per profile with idle self-close and a global concurrency cap, so a repeated check skips the login dance and a stuck browser is replaced, not babysat.
CAPTCHA solving is built in
Cloudflare Turnstile, reCAPTCHA and image challenges are solved in the flow — nobody on your team is asked to click on bicycles.
AI only where it earns its keep
When a carrier sends a return-location table as an embedded image, a vision model reads it. When a spreadsheet's headers are unfamiliar, a model maps them. Everything else is deterministic.
Notifications are change-gated
A per-run diff decides whether a report is worth sending, so a scheduled watchlist emails only when something actually moved — or once a day as a digest.
Registries everywhere
Carriers, notification channels and dashboard modules plug in through one-line registry entries. Adding a carrier changes no pipeline code.
Logins back off when a portal is unhappy
A rejected credential puts that portal on a hold that lengthens on repeat, a rate limit gets a short flat hold, and a passing probe or a saved credential lifts it — so credential rot never hammers a portal.
Dates mean one thing
ETA and last-free-day formatting happens at one choke point per module, so a BNSF date and a UP date land in your TMS in the same shape.
Get started
Give your dispatchers their day back.
A thirty-minute walkthrough on your own Chicago lanes and carriers. We will show a live trace on one of your ramps, a scheduled watchlist and the dashboard your team would open every morning.