Skip to content
roadmark

Trust center

Security and AI safety in Roadmark

Roadmark is the AI-native operating system for freight, combining a native TMS, connected applications and AI agents that prepare work for your team. Each company's data is kept apart by the database itself, not by the screens. Agents start at Draft: they prepare the work, and a person keeps or rejects it. One switch pauses every agent, and every change is recorded in a log that can't be edited.

At a glance

Price
From $499 USD/month, unlimited users
Languages
App and support in English and French
AI default
Draft: a person keeps or rejects what an agent prepares
AI off switch
Per company, per duty, a pause, and one switch for every agent
Data isolation
Row-level security, forced on every table that holds company data
Audit record
Append-only, written by the database
Sign-in
Password or single sign-on with Okta or Microsoft 365
Audit export
CSV or JSON, for any date range
With the AI off
Every screen and workflow keeps working

What an agent may do

An agent works for a person, and never past them.

Every Roadmark agent works on behalf of someone on your team, with that person's access and only the duties your team delegated to it.

An agent can

  • Read what the person who started it can read, and nothing wider
  • Prepare drafts (a quote, an email, a load update, an invoice) for a person to keep or reject
  • Act on its own only for the decisions and scope your team has granted

An agent can never

  • See another company's records
  • Act beyond the permissions of the person who delegated to it
  • Get past a dollar limit, a credit hold or a compliance block
  • Edit or delete the audit history

The rules

The rules, with their numbers.

Each rule is enforced by the Roadmark app or its database, not left to a policy document.

  • Row-level security, forcedEnabled and forced on every table that holds company data. The app connects as a database role that can't bypass it.
  • One company per transactionEvery database transaction is scoped to a single company, so no query mixes two companies' rows.
  • Delegation, capped by the personAn agent's authority is what your team delegated to it, intersected with the permissions of the person it works for.
  • Connected assistants, scopedClaude or ChatGPT connected over MCP gets the person's own grants, intersected with the scopes they consented to.
  • Suggest, Draft or ActYour team sets how far each agent goes. With no grant, an agent stays at Draft.
  • Off at every levelSwitch agents off for a company or for one duty, pause them, or stop every agent at once.
  • An audit log written by the databaseTriggers record each change: before and after values, who, on what authority, and when. The app's role can't insert, update, delete or truncate it.
  • Short sessionsPasswords are hashed with scrypt. Access tokens last 15 minutes, and refresh tokens 7 days.
  • Sealed credentialsThe keys and tokens of the systems you connect are sealed with AES-256-GCM before they're stored.

How isolation works

One database, and a scope for every company.

The group's companies keep separate data. A person's access is a scope over one company or several, and anything that runs on their behalf, an agent or an AI assistant, inherits that same scope.

How access and isolation work across the group's companies
Access and isolationAshgrove Transportation Group
  • Ashgrove Freight

    • Loads and dispatch
    • Drivers and units
    • Customer rates

    Group controller

  • Ashgrove Logistics

    • Quotes and RFQs
    • Carrier network
    • Customer rates

    Group controller

  • Keel Transport

    • Loads, on McLeod
    • Its own carriers
    • Its own invoices

    Group controllerKeel dispatcher

Each chip is one person's access. The group controller's scope spans all three companies; the Keel dispatcher's scope stops at Keel.

Every agent, and a question asked in Claude, ChatGPT or Gemini, runs as the person who called it. It carries exactly the same chips, and never more.

The database enforces it. Row-level security refuses a query that reaches outside a person's companies, whether it comes from a screen, the API, an integration or an assistant.

Sign-in and access

Every person signs in as themselves.

Each person has their own account, with a password or through single sign-on, and their access follows the role you give them at each company.

  • Single sign-on. With Okta or Microsoft 365.
  • Roles are scoped to a company. The same person can hold a different role at each company they work in.
  • Access removal. Confirm session revocation and provisioning behavior for your identity-provider setup during your security review.
Access by role
Access, by role
Access by role
RoleCompaniesCan approve
Group controllerAll companiesEverything at its level
Company adminOne companyEverything in that company
DispatcherOne companyBoard and tracking only
Billing clerkOne companyInvoices and pay only

The audit record

Every change is recorded: who, what and when.

A change always carries who made it (a person, an agent, an assistant or an integration), what changed, and the time. The database writes the record, and the app can't change or delete it.

  • Exportable. Auditors get a file for any date range, not a screen share.
  • Company by company. An export never includes another company's rows.
An audit export request
Audit export
FormatCSV or JSON
RangeAny date range you choose
IncludesWho, what changed, before and after, and where it came from

Availability

Roadmark keeps running when the AI doesn't.

Every screen, quote and load works without an agent or an assistant, whether a model provider is down or your team has paused them.

  • Works without AI. No screen or workflow depends on a model responding.
  • Backup requirements. Discuss retention, restore testing and recovery requirements for your deployment with our team.
  • Service availability. Contact us for availability information and the incident communication process for your deployment.

For buyers

Questions to ask any TMS vendor about AI and data.

Ask every vendor on your list, us included. Each question comes with Roadmark's own short answer.
  1. Does the AI act without a person's approval?

    An agent that sends a quote or books a carrier on its own commits your company to it.

    RoadmarkNot by default. Agents start at Draft, and act on their own only where your team has granted it.

  2. Can we turn the AI off, and does the system still work?

    On a bad day you need to stop the AI without stopping the business.

    RoadmarkYes. Per company, per duty, a pause, or one switch for every agent, and every screen keeps working.

  3. Is each company's data separated in the database, or only in the screens?

    A filter in the screens is one bug away from showing another company's freight.

    RoadmarkIn the database. Row-level security is forced on every table that holds company data.

  4. Is there an audit trail of what the AI did?

    When an agent changes a load or an invoice, you need to see who, what, when and on whose authority.

    RoadmarkYes. Database triggers record every change by a person or an agent, and the app can't edit the log. Export it as CSV or JSON.

  5. Can we export all our data if we leave?

    Records you can't take with you lock you in.

    RoadmarkYes. Export your records at any time in a standard format.

  6. Where is our data hosted?

    Contracts, customers and regulators can require a region.

    RoadmarkTell us your region and requirements, and we'll confirm the hosting options before you commit.

  7. Who are your subprocessors?

    Every provider that handles your data is part of your risk.

    RoadmarkThey're listed by name, from hosting and email to the driver app's push and voice providers. See the subprocessors

Data location and contact

Where your data is hosted, and who to ask.

  • Data locationTell us your hosting region and deployment requirements. We will confirm the available options and the scope of data residency, including any external processing, before you commit.
  • SubprocessorsThe providers that run the Roadmark platform, and the kinds of integrations you can connect, are listed by name. See the subprocessors
  • Security contactTo report a vulnerability, or to ask for a questionnaire or a review: hello@roadmark.ai

Questions

Questions about security and AI.

Can the AI send anything without approval?

Not unless your team allows it. Every agent starts at Draft: it prepares the quote, the email or the invoice, and a person keeps or rejects it. An agent acts on its own only for the decisions and scope your team has granted.

Can we turn the AI off?

Yes. Switch agents off for a company or for one duty, pause them, or use the one switch that stops every agent. Roadmark keeps working, and people do the same work by hand.

Who can see our data?

The people you give access to, within the companies and roles you set, and the agents and connected assistants working for them, which see only what that person can. Row-level security in the database keeps every other company out.

Is there an audit trail of what the AI did?

Yes. Database triggers record every change, whether a person, an agent, an assistant or an integration made it: before and after values, who, on what authority, and when. The app can't edit or delete the record, and you can export it as CSV or JSON for any date range.

Where is our data hosted?

Tell us your region and requirements, and we'll confirm the hosting options, backup locations and any external processing before you commit. The providers that run the platform are listed on the subprocessors page.

Can agents get past our dollar limits or credit holds?

No. An agent can't go past a dollar limit, a credit hold or a compliance block, and it can never do more than the person it works for is allowed to. Those decisions stay with a person.

Does Roadmark work without AI?

Yes. Every screen, quote, load and invoice works with the agents off, whether your team paused them or a model provider is down.

What happens to our data if we leave?

You keep an export of your records, and we delete what remains on the schedule set out in your contract.

Who do we contact for a security review or a questionnaire?

Our security team, at the address on this page. Tell us your format and deadline and we'll work to it. hello@roadmark.ai

Talk to our security team.

Send us your questionnaire, or book a call with the person who can answer it directly.

  • A straight answer, not a sales call
  • Your IT and security team are welcome
  • An NDA first, if you need one

Straight to our security team, not sales.

We use this to reply to your request. Privacy policy.