Pick the AI model and provider for every application and agent from one shared registry, with a single default for the whole estate. A model still assigned to something cannot be deleted. You see cost for text and images live on screen while a run is still going.
The Agentic Part
An agent tests every connected service at the press of a button and reports immediately what is wrong, without hanging on a slow response. For every image generation, a second agent picks the cheaper of two vendors on its own, with no change needed on the calling application’s side. A third ties every processing stage of a run together under one ID and pulls the cost straight from the provider’s own billing data.
Main Features
- One shared registry of AI models for every application and agent
- One catalogue-wide default, changed in a single place
- Provider, context window, pricing and image support recorded per model
- A model still assigned to an agent cannot be deleted
- One overview screen with every connected service and its detail
- Genuine test button, separate from simply toggling a connection
- Test that never hangs, even when a service is slow to answer
- Second vendor on standby, no change to the calling application
- Price list per model with actual per-image cost recorded
- Spend table separating pass-through from absorbed cost
- Every charge recorded exactly once, even on retry
- Dedicated log line per processing step, tied by one ID
- Cost capture for text and images from provider billing data
- Live progress and cost feed to the screen
- Written run record for after-the-fact troubleshooting
Use Cases
- Every application used to pick its own AI model, with prices and providers scattered across several places. Now every application and agent draws its model from the same registry, with one default for the whole estate.
- A connection to an AI provider goes down. The test button shows exactly why, before the live operation is affected.
- Two vendors are available for an image generation job. An agent picks the cheaper one automatically, with no change to the calling application.
- A single run, such as a scan or a generated article, needs to be traceable end to end later. Every processing stage writes its own log line under the same ID.
What we build for you
- Which models should be available to which applications
- Who is allowed to change the catalogue-wide default
- How each service communicates, and what counts as connected
- Who on your team may change these connections
- Your own vendor accounts and the project
- Which steps your staff should see, and where the records are kept.
