Platform · AI and agents
AI-ready vehicle data with permissions built in.
Ask your fleet a question in your own model names and get the answer in seconds: no export, no data team, no week of waiting. AI projects fail on their data more often than on their model, so every answer comes from one clean record per vehicle, and only from the vehicles the asker may see. Today that runs over our query and data APIs. Next, Claude or ChatGPT ask the same questions.
Product manager · Coride Motors
Distance and error codes per model family, month by month, July to September.
- No export
- No SQL
- No data team
AnswerVehicles you built, and only those
| Model family | Month | Distance | Error codes |
|---|---|---|---|
| Urban | Jul | 412,880 km | 41 |
| Urban | Aug | 398,215 km | 36 |
| Urban | Sep | 377,090 km | 52 |
| Touring | Jul | 188,340 km | 9 |
| Touring | Aug | 201,775 km | 12 |
| Touring | Sep | 163,410 km | 8 |
At a glanceUrban: error codes up 44 % in September, while distance fell 5 %. Worth a look before the season ends.
Insight query · one question, one answer · demo data
01Live and next
What runs today, and what comes next.
Kept apart, because AI claims are easy to make and hard to check.
Live today
Insight queryLive
Your team answers its own questions.
Distance, energy, error codes and more, per day, week, month or year, split by your own model range.
Vehicle dataLive
Train your own models.
Any fields for any vehicle you may see, as CSV or JSON.
Live streamLive
Your systems react the moment a vehicle does.
Status, actions and notifications over WebSocket.
Service projectionLive
Workshops plan before the owner calls.
Weeks to the next service, from each vehicle's own usage.
Trip validationLive
Honest numbers.
A trained classifier separates real trips from noise.
Next
AI toolboxComing
Claude or ChatGPT ask your fleet.
An MCP connector with two read-only tools: one lists what can be asked, one asks it. In development.
Sign-in for agentsComing
An agent is never more than its person.
An agent signs in as a person and may only read insights.
Vehicle assistantComing
Owners get help in your app, any hour.
Answers from the vehicle's status, its open errors and your troubleshooting texts, and opens a ticket. Shown as a demo today.
02Vocabulary
What a question can ask for.
Pick a measure, a period and a split. The same words in the API, in the OEM Suite and, next, in the AI toolbox.
Ask for · 12 measures
- Distance
- Time on the road
- Average speed
- Top speed
- Trips
- Active vehicles
- Fuel used
- Fuel consumption
- Energy used
- Energy per km
- CO₂ saved
- Error codes
Per period
- Day
- Week
- Month
- Year
- In total
Split by
- Vehicle
- Vehicle class
- Drivetrain
- Country
- Your categories
- Your model families
- Your models
Makes“Error codes, per month, by model family.” Answered in your model names, not ours.
Only measures your vehicles can produce are offered: an electric boat maker is never asked about fuel, a tractor maker never about CO₂ saved.
03Permissions
No answer shows more than you may see.
A question goes through the same checks as every screen. An agent will see what its user may see, and nothing more.
As the manufacturer
Every vehicle you built, across markets and partners.
As a partner
The vehicles a dealer, workshop or fleet looks after, and no others.
Per request
The permissions of the person asking, resolved again for every call. Each answer says which permission let it through.
Why you can trust an agent with it
An agent never writes its own database queries. It can only ask the questions in the catalogue, through the one path every screen uses.
So even an agent tricked by a clever prompt cannot reach another organisation's vehicles, or anything outside what its person may see.
04Questions
About AI on the platform.
Can we train our own models on the data?
Yes. The data API returns any fields for any vehicle you may see, per period, as CSV or JSON, and the query API returns aggregates over your whole fleet. Both carry the permissions the data was collected under.
Can Claude or ChatGPT ask questions about our fleet?
That is the AI toolbox, an MCP connector, and it is in development. It exposes the insight query as two read-only tools behind OAuth 2.1, so an agent signs in as a person and sees what that person may see.
Can a prompt trick an agent into showing more?
No. An agent never writes its own database queries; it can only ask the questions in the catalogue, through the same path every screen uses, with the permissions of its person. A manipulated prompt changes the question, never what may be answered.
Is this predictive maintenance?
Partly, and we say which part. Service due dates are projected from each vehicle's own usage trend, in weeks to the next service, and a notification goes out before it is due. It is a projection from real usage, not a black box.
Does the vehicle assistant ship in our app?
Not yet. It runs as a demo: it reads the vehicle's status, its open errors with your troubleshooting texts, its maintenance plan and parts, and opens a support ticket, by text or by voice. It is not in the production apps yet.
Where to go next