Skip to main content
The OpenAI Codex agent runtime in Mains communicates with the Codex app server via JSON-RPC over stdio. These settings control sandbox isolation, network access, code review, and more.

​
Sandbox Mode

Controls file and network isolation for the Codex agent during runs.
Full Access mode disables all file isolation. Only use this in trusted environments where the agent should have unrestricted access to the filesystem.

​
Network Access

Toggle network access within workspace-write sandbox mode. When enabled, the agent can make outbound network requests during runs. Disabled by default in read-only mode. Allow the agent to search the web during runs. When enabled, the agent can fetch live search results to inform its responses and code generation.

​
Skip Git Check

By default, Codex requires the workspace directory to be a Git repository. Enable this option to allow running in non-git directories.

​
Code Review

Codex supports a native review mode. You can request a structured code review that analyzes workspace changes and produces findings with severity ratings. Review targets: Findings are parsed from the Codex output with priority levels (P1 = critical, P2 = warning, P3 = info) and saved as structured review findings.

​
Session Management

Codex runs support the full session lifecycle:
  • Continue — resume a previous session to pick up where you left off

​
Account & Rate Limits

View your Codex account information and current rate limit status from the settings panel. Rate limits are retrieved from the Codex app server and displayed per-model.