Security
Companies stay apart. The rest is a plain list.
Several courier firms share one system, so the first thing to get right is that they cannot see each other. This page says what we have done and, further down, what we have not.
1Isolation
One database, separate drawers.
Your rows and another company’s rows sit in the same tables, and the database decides which ones a request may see. If we wrote a bug in an endpoint, the database would still refuse to hand over another company’s data. That is the point of doing it there rather than only in application code.
2Who sees what
Four roles.
Accounts are created by an admin. The company account signs in with the company number. Drivers and dispatchers sign in with their email. Everyone has a password. There are no invitation links.
- A driver
- Their own routes and their own history. Not other drivers, not other companies.
- A dispatcher
- Their company's routes, stops and drivers. Cannot manage users.
- A company admin
- Everything in their own company, including accounts and zones.
- The platform operator (super admin)
- Creates and suspends companies and sees usage and system health. Can act as a company when support needs it, in a session that lasts an hour and is written to an audit log.
3In place
What is done.
Companies are kept apart in the database
Every table that holds a company's data carries the company, and the database itself refuses to return another company's rows (row-level security). It does not rely on each request remembering to filter.
Passwords are hashed
With argon2id, never stored as typed, never written to logs. Minimum length is eight characters. Admins can reset a password and people can change their own.
Sessions
On the web, a cookie that scripts on the page cannot read, sent only over HTTPS, and tied to the site. In the app, short-lived access tokens with refresh tokens that rotate, kept in the phone's secure storage. Signing out revokes them.
Roles are checked on the server
Hiding a button is not security, so every request is checked against the person's role and the company the record belongs to. Anything not explicitly allowed is refused.
Sign-in and public endpoints are rate limited
Repeated attempts are limited, so guessing passwords gets slow and then stops.
Uploads go through signed links
Manifests and photos are uploaded to object storage on our own server using links that expire, not through a public folder.
4Other people’s servers
Where data leaves our server.
Reading messy files. When a language model is switched on to read a photo or a messy file, the contents of that file are sent to a model provider through OpenRouter. When it is off, or unavailable, a simpler rule-based reader on our own server does the job with less skill. Either way, a person checks the result.
Map tiles. The pictures of the map in the browser and the app are loaded from a public tile service built on OpenStreetMap data. Addresses and routes are not sent there, only the requests for map squares.
Address search, road distances and route ordering all run on our own servers.
The honest bit
What is not done.
We would rather you read this here than discover it later.
Certifications
No SOC 2, no ISO 27001. We are not going to say we are compliant with something we have not been audited against.
An independent security test
Not done yet. We have reviewed our own code more than once, which is not the same thing.
Monitoring and alerting
Error tracking, uptime monitoring and alerts are part of the work before real customers, and are not finished.
Encrypted storage on the phone
The app's local database leans on the phone's own protection. Encrypting it ourselves is planned.
A track record
In pilot, nothing has been through years of real use. Treat it accordingly and start with data you can live without.
Found a hole, or want to ask how something works? Write to hello@routeplanner.example. Or read how we handle the data you give us on this site.