Agenten-Kompass

AGENT PROFILES

Codex

OpenAI · Coding agents

OpenAI's agent for software work: understand, change, and test code. Available in the desktop app, terminal (CLI), editor, and cloud.

Develop software

ChatGPT and Codex · Coding agent

Product and access

Codex is OpenAI's agent for software work. You can use it in the desktop app, in the terminal through the Codex CLI, in the code editor, or in the cloud. The CLI provides command-line access. dots can delegate development tasks to Codex.

01

Core features

10 features from provider sources · sources reviewed on September 30, 2026. Availability and limitations are listed directly below.

  • Inspect and modify code

    Codex handles development tasks based on a project, can prepare changes, and run the associated checks.

    Source 1

  • CLI, IDE, and cloud

    You can use Codex in the terminal, alongside code in the editor, and in configured cloud environments. The access method affects the workflow.

    Source 2

  • Parallel work in worktrees

    Independent tasks can use separate Git working copies. Changes and files remain separate for each worktree.

    Source 3

  • Move tasks between workspaces

    Handoff transfers a task between a local checkout and a managed worktree. Git state and ignored files must be considered.

    Source 3

  • Skills for project workflows

    Skills bundle instructions and resources for recurring work. They can describe project routines such as reviews or documentation.

    Source 4

  • Scheduled development tasks

    You can schedule tasks in the desktop app. Git projects can use dedicated working copies for this. The computer and app must be running for local work.

    Source 9

  • Project rules with AGENTS.md

    Global and project-specific instructions define working methods, conventions, and checks for development tasks.

    Source 5

  • Specialized subagents

    Independent subtasks can be assigned to parallel agents. Local Codex clients support custom agent configurations; additional agents consume additional usage.

    Source 6

  • Connect MCP tools

    Local Codex clients connect to local or remote MCP servers. App, CLI, and IDE share the configuration of the same host.

    Source 7

  • Hooks for your own checks

    Trusted hooks run scripts or MCP tools on agent events, for example for validation, logging, or policy enforcement.

    Source 8
02

Requirements and limits

‘Still open’ means the reviewed sources do not answer this point clearly.

  • Requirement

    Access to files and the network

    The environment and permission profile determine which files Codex can read or change and which services it can access.

    Source 10

  • Requirement

    Local schedules require the app to be running

    For scheduled work in a local project, the computer and desktop app must be running and the project available.

    Source 9

  • Limitation

    Schedules are managed in the app or on the web

    The CLI and editor extension do not have their own Scheduled management. You can prepare tasks there; they are configured in the app or on the web.

    Source 9

How you can use it

Desktop app

Shared workspace for ChatGPT, Codex, and dots. Local files and computer tools require appropriate access permissions.

Provider documentation

Editor extension

Use Codex directly beside the code. The editor is another way to access the coding agent.

Provider documentation

Custom client via app server

Connect a custom interface to Codex: present conversations, live events, and approvals in your own product.

Provider documentation

Desktop app in detail →

Codex App Server: your own client

The app server connects your interface to Codex: sign-in, conversation history, approvals, and live events. For automated jobs and CI, the documentation recommends the Codex SDK.

The app server command and WebSocket transport are documented as experimental and unsupported for production workloads. Check the specific interface and version before integration.

Provider documentation
03

Practical tasks

Editorial examples based on documented features. No product tests performed.

01

Add a small feature

Your input
Repository, acceptance criteria, and existing project rules.
Possible result
A verifiable diff with suitable tests and notes on remaining questions.
What to check
Test the feature in the actual workflow and review the code.
02

Prepare two independent changes

Your input
Two clearly separated tasks in the same Git project.
Possible result
Separate worktree changes for review and later integration.
What to check
Check interactions and merge conflicts before adoption.
EDITORIAL EXAMPLE

Implement a limited feature change in a repository and review the diff.

Your possible prompt

Read the project rules. Investigate [error], propose a small change, and run the relevant checks. Show me the diff and any remaining uncertainties.

Getting-started idea based on documented capabilities. Not an executed product test. Adapt the task, data, and permissions.

04

What you should know first

This profile describes development work. App features such as document views or general Work tasks are covered in the separate desktop app profile.

Requirements & limitations

Review code, tests, and changes. Specific access depends on the environment, plan, and project configuration.

Setup

Provide project rules and the task.

Pricing model

Plan / usage. Check current terms and included quotas with the provider.

The classification indicates the scope of effort. It is not a price or performance commitment.

05

Sources from the provider

Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.

Read profile as Markdown

How we assess the sources