artiplane
Demo
Pillar 01 · Cost governance

From prompt to response, through six checks

Artiplane sits between people and models. Every request is classified, checked, routed to the most suitable provider, verified and billed at real cost.

01 · Request

The user writes in the company chat. No configuration, no manual model choice.

02 · Classification

The request is classified by category, purpose (work or personal) and risk level under the AI Act.

03 · Control gate

Area permissions, personal data detection, company policies and protection against instruction-bypass attempts.

04 · Routing

Selection of the cheapest model that meets the task, within the area's hosting, context and latency constraints.

05 · Verification

The response is checked against the task's criteria, with automatic escalation to a higher-tier model if quality is insufficient.

06 · Accounting

Real cost recorded in micro-euros and attributed to user and area, with the balance updated right after the response.

What AI routing decides on

The router doesn't pick "the best model": it picks the cheapest model that solves the task while respecting the area's constraints. Five criteria, applied to every single request.

Task category
Summarisation, translation, legal analysis, code, sourced research, vision: each category has a dedicated pool of models.
Confidentiality constraints
An area handling sensitive data can be restricted to providers hosted in the European Union only.
Available budget
The cost estimate is checked against the user's and the area's balance before sending: without cover the call doesn't go out.
Provider status
Outages, rate limits and out-of-threshold latency trigger a fallback to an equivalent model.
Personal use
Questions unrelated to work are routed to a dedicated lightweight model, separate from the company knowledge base.

What the administrator sees

Spend in euros

Spend by area, user and use case, with the gap between estimated and charged cost.

Areas and budgets

Company, area and individual budgets kept consistent, with alert thresholds and top-up workflows.

Interaction log

Every prompt with a SHA-256 fingerprint, model used, risk classification and outcome.

What changes for people at work

  • A single interface for all models, with no provider to choose
  • Request cost and remaining balance visible right after the response
  • Company policies accessible with one click from the chat
  • Free personal questions, kept separate from the work context
  • Documents and presentations generated with the company template
  • No interruption when a provider has an outage

Learn more: how multi-provider AI routing works and how the cost per response is calculated.

Want to see Artiplane on your own use cases?

A 30-minute demo on your real scenario: routing, budget per business area and full interaction tracking.

Request a demo