The request is sent
The team connects its editor or application to the gateway and uses mattear-auto to delegate model selection.
Mattear · Governance and AI FinOps
Every task,better served.
AI Management brings together model selection, prompt evaluation, and spend control. With mattear-auto, Jev seeks the right model for each request at the lowest reasonable cost, from the models your organization has enabled.
Automatic selection · Evaluation with Jev · Budgets and traceability
Illustrative example · Not actual results
Your team asks
“Write tests for this function and cover the edge cases.”
Jev selects from your enabled modelsTask capability + reasonable cost
After the responseCost recorded. Prompt evaluated.
Quality, task type, and estimated savings against your reference model.
How the gateway works
Select, control, learn.
Your team keeps its compatible tools. The gateway connects each request to available models and brings together the signals to improve AI usage.
Jev selects the model
It considers the task, capabilities, and prices of enabled models compatible with the request.
Your controls apply
Policies and budgets can block or change the selection before the request runs.
Everything stays visible
The cost is recorded alongside estimated savings when available. With the evaluator enabled, Jev analyzes the prompt in the background.
mattear-auto
The right model. A cost that makes sense.
Every task has different demands. Jev uses your organization’s context to seek sufficient quality at the lowest reasonable cost, without always choosing the most expensive or the cheapest model.
Your catalog, your rules
Only from enabled models.
Selection considers active models with configured access and current prices. It filters for compatibility with tools, streaming, images, and request limits before choosing.
The administrator enables mattear-auto for the organization. The team uses it like any other model from compatible chat clients, including VS Code, Zed, and custom applications.
Savings with context
Compare each decision with a baseline.
Activity shows the applied model and estimated savings per request, with a total for the selected period and filters.
The estimate compares the cost of the same tokens against your reference model and subtracts the cost of the model used and the selector. It can be negative: it is neither a savings guarantee nor a comparison of equivalent quality.
Evaluator powered by Jev · TypeSafe
Better requests start with better signals.
Beyond selecting models, Jev evaluates prompts after the response. Your team gets concrete criteria to review how it asks AI for help, without adding that evaluation to response time.
Prompt quality
Clarity, context, instructions, and expected outcome, with a score from 0 to 100 and a traffic-light indicator in Activity.
Task type
Configurable categories to understand what AI is asked to do: development, review, testing, documentation, and more.
Your company’s context
Signals of possible misalignment with the organization’s defined context, for human review.
History review
Reevaluate between 1 and 90 days using your current criteria and review model recommendations for future requests.
Evaluation analyzes the prompt, not response quality or individual performance. Its signals do not block requests or confirm misuse.
Key capabilities
Governance without slowing adoption.
Finance gains visibility, technology keeps its flexibility, and every team understands what it consumes and why.
Budget by team
Limits by organization, area, and person, with real-time control and automatic blocking.
Exception workflow
Fast, auditable approvals when a team needs to exceed its limit.
Provider flexibility
Switch providers without changing the developer experience or tools.
Automatic selection
mattear-auto seeks the right enabled model based on task, capabilities, and cost.
Evaluation with Jev
Prompt quality, task classification, and alignment signals based on your organization’s context.
Attribution by project
Accurate cost allocation for finance, chargeback, and data-driven decisions.
Governance and security
Keys and policies in one place.
Centralized keys, role-based access, auditing, traceability, and operational continuity when providers change, plus an executive cost dashboard.
Delivery options
SaaS or within your infrastructure.
Available as a service priced per created user, or deployed in the client’s cloud to keep data within their environment.
Frequently asked questions
Which usage can it control?
The gateway attributes the usage routed through it to organizations, teams, developers, and projects. To include an application, its requests must go through the gateway.
Does mattear-auto always choose the cheapest model?
No. It seeks a model capable of handling the task with sufficient quality at the lowest reasonable cost. It only considers enabled models with current prices that are compatible with the request. Policies and budgets still apply.
What is needed to enable Jev and mattear-auto?
The administrator configures a TypeSafe token with access to Jev and the organization’s context. For mattear-auto, they also define a reference model and the selector rate. Both features start disabled; TypeSafe usage is billed separately.
What information does the evaluator receive?
It receives the user’s request extracted from the payload and the configured context. It does not receive the user’s email or the model’s response. Evaluation requires a recorded, identifiable request.
Are displayed savings guaranteed?
No. It is an estimate against the reference model for the same tokens, minus the cost of the applied model and the selector. It can be negative and is not calculated for failed requests or requests without known prices.
What deployment options are available?
It is available as SaaS or deployed in the client’s cloud. During the assessment we review providers, teams, and usage patterns to choose the deployment model.
How is the service priced?
The SaaS option is priced per created user. The scope of deployment in the client’s infrastructure is defined in the proposal based on their operation and requirements.
Next step
Find where your AI can deliver more.
Let’s build a pilot with your actual models and tasks. We review prompt quality, try mattear-auto, and compare costs against an agreed baseline.
- ✓ Consumption assessment
- ✓ Pilot with mattear-auto and Jev
- ✓ Cost comparison and opportunities