---
title: "What Is an Agent Console? Understanding Work and Control Properly"
description: "Agent Console explained simply: how work sessions become visible and why policies, memory, and risk-based approvals belong in their own control center."
type: "wissen"
product: "workspace"
slug: "what-is-an-agent-console"
language: "en"
source_id: "wissen/was-ist-eine-agent-console"
published: "2026-06-10"
status: "publish"
faq_json:
  - q: "What is an Agent Console?"
    a: "An Agent Console makes the work environment of an agent accessible. Depending on the task, it displays sessions, runtime environments, inquiries, and work results, so that delegated work remains traceable."
  - q: "How does an Agent Console differ from a chat?"
    a: "A chat organizes messages. An Agent Console organizes ongoing work with state, work environment, and handovers. This allows the agent to work over the course of a longer assignment rather than simply delivering individual responses."
  - q: "Is live observation the same as control?"
    a: "No. Visibility helps with diagnosis and handover. Control emerges through controllability: clear mandates, policies, limited permissions, memory, and approvals for risky steps."
  - q: "What role does webRichtung workspace play?"
    a: "workspace is a browser cockpit available on request for agent and coding sessions. It manages sessions, runtimes, hosts, and approvals around these work environments."
  - q: "Where is agentic work controlled across the enterprise?"
    a: "The agent module is the control center for assignments, automations, policies, risk-based approvals, heartbeats, memory, and integrations."
---

An Agent Console is an interface for delegated, ongoing work. It displays not just a dialogue but the state of a work session, its environment, and pending handovers. This is especially important where an agent edits files, modifies code, or executes other multi-step tasks. A console makes this work accessible—but it is not automatically the entire control system of the enterprise.

## Why a chat doesn't capture every kind of work

A chat organizes messages in sequence. Agentic work has an additional dimension: a state. An assignment can be running, waiting for context, requiring a decision, or delivering a result for acceptance. Furthermore, an agent may work in its own runtime environment with files and tools.

An Agent Console makes this work context visible. You can see which assignment lies in which session and where a handover is required. The agent gets a workspace for execution; you get traceable access to this work.

## Typical components of a work console

For agent and coding work, these areas are particularly relevant:

- **Sessions:** separate work units with a discernible state;
- **Runtimes:** the environments where the work is actually executed;
- **Hosts:** the assigned systems behind these environments;
- **Interaction:** a way for inquiries, additions, and deliberate interventions;
- **Approvals:** clearly marked points where a risky step does not proceed independently.

Which components are necessary depends on the task area. A telephone reception does not need a coding terminal; a coding session, by contrast, needs a clearly isolated work environment.

## Visibility is not the same as control

It is tempting to equate control with permanent live observation. Yet an agent does not become reliable simply because you can see every intermediate step. Effective control begins earlier: with a clear mandate, appropriate permissions, policies, and boundaries.

Control means controllability. Within a safe framework, technologically generated work capacity should be able to operate independently. Human approval is required where a step is risky or exceeds the authority granted—not blanket approval for every action.

## workspace: Cockpit for coding sessions

[webRichtung workspace](https://www.webrichtung.de/en/modules/workspace/) is a browser cockpit for agent and coding sessions. It organizes sessions, runtimes, hosts, and associated approvals. The focus is on genuine work environments for SSH, tmux, and Codex-based sessions, not on universal agent control for every business area.

workspace is available on request. This distinction is important because such work environments bring their own access and security considerations. Getting started should match the specific assignment and the required runtime.

## agent: the actual control center

Enterprise-wide agentic work is controlled in [webRichtung agent](https://www.webrichtung.de/en/modules/agent/). Assignments, automations, policies, heartbeats, memory, integrations, and risk-based approvals come together there. Memory keeps relevant context available; policies define permitted behavior; approvals mark the steps where a human must decide.

The separation is useful: workspace provides a workspace to selected agents, agent holds mandate and control together. Both can be part of the same operating system without claiming to do the same thing.

## What you should clarify before deployment

Define the work assignment first, and the interface only after. What environment does the agent need? Which data and tools are permitted? What results should it deliver? Which steps are risky? Who decides in an escalation?

Once these questions are answered, the console helps with execution and handover. Without this framework, even complete visibility remains mere observation. How an agent's mandate is structured is explained in [What Is a KI Agent?](/en/knowledge/what-is-an-ai-agent.html).