Control surfaces for AI agents

An agent control surface shows what agents are doing, which sources and tools they use, what is uncertain, and where humans need to intervene.

What this is for

This is for teams that need to monitor agent runs, inspect errors, see costs, and route open decisions back to people.

  • An operator UI makes agent work visible: status, tool use, sources, costs, errors, retries, and intervention points live in one place.

  • appamass does not think about agents without their interface: Google Agent Development Kit (ADK) and Python run the agent; React/Vite and TypeScript make it operable and auditable.

  • The first step is a readable run history with steps, result, errors, and the point where a person can take over.

Organize runs, tool traces, and escalations

AI agents need a workspace for humans. It has to show what happened, what is waiting, and which decision should remain manual.

What users see

Users see run history, tool calls, sources, cost, open decisions, retries, and manual interventions.

How it works

React/Vite, TanStack Query, Agent Development Kit (ADK) run models, filters, audit logs, observability, cloud logs, backend contracts, and roles sit behind the surface.

What stays controlled

Rights, cost, error classes, retries, manual overrides, approvals, and agent-run evaluation stay controlled.

A first agent cockpit

The first build should make the most important agent runs visible and give the team real intervention points.

Show run history

Each run shows status, input, intermediate steps, result, error, and duration.

Review tool activity

Tool calls, sources, and data access are traceable before a result is reused.

Route escalation

Open cases go to the right role with context, priority, and a clear action.

Related areas showing how mobile apps, React web systems, AI agents, and controllable automations fit together.

Start project or scale further?

We support you.