My own product · In private pilot
One AI connection. Every client system.
A managed AI service that gives the systems I build one connection to AI models, with limits and billing for each customer.
Modelvia is the AI layer behind the software I build. Instead of each system holding its own AI accounts, they connect once to Modelvia. It picks a suitable model for each request, keeps every customer inside their own spending cap, and records the cost of each request.
Visit modelvia.dev (opens in a new tab)
No tech knowledge needed · 15 minutes · you keep the map either way.
- Best for
- Businesses running AI inside their own software that need costs and limits for each customer.
- Problem
- Each AI system needs its own model accounts, spending limits and usage records, and they drift apart.
- Changed
- One connection with automatic model choice, a cap for each customer and a cost record for every request.
The plumbing every AI system repeats.
Every system that uses AI needs the same groundwork: accounts with model providers, a way to stop one customer running up costs, and records accurate enough to invoice from.
Built separately into each system, that work is repeated and easy to get wrong. A missed limit or a missing record turns into an unexpected bill.
How a request travels.
Connect once
A system points its existing OpenAI-style code at Modelvia by changing the address and the key.
The model is chosen
Modelvia picks a model for each request, fast and inexpensive or more capable, unless the system asks for one.
Every request is accounted for
Each customer has their own project, keys and monthly cap. Requests over the cap are refused before they run, and each one leaves a cost receipt.
The usage portal.
Each customer project sits against its own cap, and every request can be traced to a receipt.
What it does.
It is the shared service for the applications I build, so each new client system starts with limits and records in place.
One connection
Accepts standard OpenAI-style and Anthropic-style requests, so existing code needs only a new address and key.
Automatic model choice
Three modes: auto, flash for fast and inexpensive work, and max for the most capable model.
Limits for each customer
Each customer gets their own project, keys and monthly spending cap.
A receipt for every request
The cost of each request can be looked up afterwards, so an invoice traces back to real usage.
A portal for the numbers
Signed-in pages for usage, keys, models, team and billing.
Backup and recovery
The usage ledger is backed up encrypted, and a restore has been tested.
Where it stands.
- Status
- In private pilot; public signup is closed
- Payments
- Customer payment collection is not switched on yet
- Runs on
- A hosted web service, API and portal
- My role
- Product design, engineering and operation
How costs and data are kept in check.
Spending caps are checked before a request runs, not after. Requests are routed to model providers selected for zero data retention. Every request leaves a cost record, so an invoice can be checked line by line.
How it is built.
- Gateway
- A hosted Node service with one writer for the ledger
- Ledger
- PostgreSQL, with encrypted backups
- Portal
- A Next.js website for usage, keys and billing
- Compatibility
- OpenAI-style and Anthropic-style requests
Building AI into your own software?
Modelvia shows how I approach the unglamorous parts: limits before costs, records before invoices. Tell me what your system needs to do.
No tech knowledge needed · 15 minutes · you keep the map either way.