Operations panel: decisions, agents and AI costs
One place where the owner sees the agents' work, answers their questions with one tap and keeps the cost of language models in check.
A recording and screenshots of the panel running on sample data: made-up agent questions to the owner and made-up model calls with amounts, no connection to the database or network; missing Polish interface labels translated, the finance bar hidden in the recording.
Task
Agent work, alerts and model costs were scattered across several tools, and questions that needed a human decision got lost among notifications.
Solution
A FastAPI web app with server-side templates and live updates over WebSocket. Agent questions arrive as cards with options, the recommended one starred; the choice goes to the decision log and the agent carries it out on its own. The models page combines recent calls, paid account balances and a quality score for each model. On a phone the panel works as a PWA with a bottom tab bar.
Result
- A decision for an agent with one tap, also from a phone
- Model costs and account balances visible in real time
- Visual check on phone and desktop before every change
My role
Interface design, server and frontend code, automated visual tests.