OpenAI Codex CLI Settings Generator

Core & Model Settings

Configure the default AI model, provider, reasoning effort, and web search behavior.

Model

The default AI model to use for the agent.

Model Provider

The backend provider serving the model.

OSS Provider

Default local provider when using OSS mode (e.g. ollama, lmstudio).

Service Tier

Preferred service tier for new turns (flex for cost savings, fast for lower latency).

FlexCost-optimized tier.

FastLow-latency tier.

Personality

Default communication style for supported models.

NoneNo personality style applied.

FriendlyWarm and conversational tone.

PragmaticDirect and efficient tone.

Reasoning Effort

Controls how deeply the model thinks about a task (supported by GPT-5/o3 families).

Low

Medium

High

Extra High

Plan Mode Reasoning Effort

Reasoning effort used specifically during planning steps.

Low

Medium

High

Extra High

Model Verbosity

Controls Responses API text output length and verbosity for GPT-5 models.

Low

Medium

High

Reasoning Summary

Controls whether to show, hide, or compress the model's step-by-step reasoning summary.

Auto

Concise

Detailed

Model Supports Reasoning SummariesForce-enable reasoning summary capability for the current model.

Model Context Window

Override the context window size in tokens for the selected model.

Auto Compact Token Limit

Token threshold at which Codex automatically compacts conversation context.

Tool Output Token Limit

Maximum tokens allowed in a single tool output.

Model Catalog JSON

Optional absolute path to a custom JSON model catalog.

Default Profile

The default profile to load automatically on startup if not passed via CLI flag.

Web Search

Controls the live web search capability for the agent.

LiveAlways search the web in real time.

CachedUse cached web results when available.

DisabledDisable web search entirely.

Developer Instructions

Custom developer-level instructions prepended to the system prompt.

Model Instructions File

Path to a text file containing custom instructions for the model.

Notification Command

External command to run for desktop notifications (array of command and args).

Sandbox, Security & Approvals

Control when the agent pauses for approval and what it can access.

Approval Policy

Controls when Codex pauses for human approval before executing actions.

On RequestPause for approval on each action request.

UntrustedOnly pause for untrusted operations.

NeverNever pause for approval (fully autonomous).

Sandbox Mode

OS-enforced technical boundaries for what the agent can touch or access.

Read-OnlyAgent can only read files, no write access.

Workspace WriteAgent can write within the workspace directory.

Full Access (Danger)Agent has unrestricted file system access.

Allow Login ShellAllows shell-based tools to use login-shell semantics.

Authentication & Login

Configure credential storage, ChatGPT backend, and MCP OAuth settings.

Credentials Store

Where credentials (auth.json) are stored (OS Keychain vs. file system).

KeyringOS Keychain (secure).

FileFile system storage.

ChatGPT Base URL

Advanced override for the ChatGPT backend API URL.

MCP OAuth Credentials Store

Where MCP OAuth credentials are stored.

AutoAutomatically select storage.

KeyringOS Keychain.

FileFile system.

MCP OAuth Callback Port

Port for the MCP OAuth callback listener.

MCP OAuth Callback URL

Custom OAuth redirect URI for MCP authentication.

Application & TUI Settings

Terminal UI, notifications, editor integration, and telemetry settings.

Alternate Screen

Controls the alternate screen mode for the Terminal UI.

AutoAutomatically decide based on terminal.

NeverNever use alternate screen mode.

TUI NotificationsEnable or disable terminal UI notifications.

TUI AnimationsEnable or disable ASCII animations in the terminal UI.

Show TooltipsDisplay onboarding tooltips in the terminal UI.

Theme

Color theme for the terminal UI.

File Opener

Editor to open when clicking citation links.

Show Raw Agent ReasoningSurfaces raw chain-of-thought events in the terminal/logs when available.

Hide Agent ReasoningSuppress reasoning output from the agent entirely.

Disable Paste BurstDisable paste-burst detection in the terminal input.

AnalyticsEnable or disable telemetry/analytics.

FeedbackAllow /feedback submissions from the CLI.

Check for UpdatesChecks for Codex updates on startup via npm/brew.

Background Terminal Max Timeout (ms)

Max timeout in milliseconds for background terminal polling.

Project Documentation Controls

Configure AGENTS.md and project documentation settings.

Project Doc Max Bytes

Maximum bytes to read from AGENTS.md or similar project documentation files.

Project Doc Fallback Filenames

Fallback filenames to search for project guidance when AGENTS.md is not found.

Project Root Markers

Directory markers used to detect the project root.

.git

History

Configure session transcript persistence.

Persistence

Controls how session history is saved.

Save AllSave all session transcripts.

NoneDo not save session history.

Max Bytes

Maximum file size cap for history files.

Sandbox Writable Roots

Fine-grained controls for the workspace-write sandbox mode.

Network AccessAllows the agent network access to run API calls or package installers.

Writable Roots

Extra directories (outside the active workspace) the agent is allowed to write to.

Exclude TMPDIR Env VarExcludes the user's $TMPDIR environment variable from writable roots.

Exclude /tmpExcludes the system /tmp folder from writable roots.

Shell Environment Policy

Control which host shell environment variables Codex inherits.

Inherit

Controls which host shell environment variables Codex inherits.

CoreInherit core variables only.

AllInherit all environment variables.

NoneInherit no environment variables.

Include Only

Specific environment variables to explicitly allow (e.g., PATH, HOME).

Exclude

Environment variable patterns to exclude (array of globs).

Ignore Default ExcludesDisables the default exclusion filter for KEY/SECRET/TOKEN variables.

Use Profile Script (Experimental)Forces shell commands to be executed through a user profile script.

Model Provider Configuration

Configure alternative or open-source model endpoints.

Provider Name

The display name for the custom provider.

Base URL

The API base URL for the custom provider.

Wire API

API protocol to use when communicating with the provider.

Environment Variable Key

The environment variable containing the provider's API key.

API Key Instructions

Optional setup guidance displayed to users if the key is missing.

Requires OpenAI AuthWhether this provider uses OpenAI authentication (for data-residency endpoints).

Supports WebSocketsWhether the provider supports WebSocket connections.

Request Max Retries

Max network retries for standard API requests.

Stream Max Retries

Max network retries for streaming endpoints.

Stream Idle Timeout (ms)

Timeout for idle streams in milliseconds.

MCP Servers

Model Context Protocol (MCP) server integrations.

Command (stdio)

Launcher command for an MCP stdio server (e.g., "npx").

Arguments (stdio)

Arguments for the launcher command.

Environment Variables (stdio)

Additional environment variables set for the stdio server.

Forwarded Env Vars (stdio)

Environment variable names to allow/forward to the stdio server.

Working Directory (stdio)

Custom working directory for the MCP stdio server process.

URL (http)

Server address for a streamable HTTP MCP server.

Bearer Token Env Var (http)

Env var sourcing the bearer token for an MCP HTTP server.

HTTP Headers (http)

Static header name-value pairs for the MCP HTTP server.

Env HTTP Headers (http)

Header names mapped to environment variables for the MCP HTTP server.

EnabledEnable or disable the MCP server.

RequiredFail Codex startup if this MCP server is unavailable.

Startup Timeout (sec)

Timeout in seconds for the MCP server to start up.

Tool Timeout (sec)

Timeout in seconds for individual tool executions.

Enabled Tools

Allowlist of tool names the MCP server is permitted to expose.

Disabled Tools

Denylist applied after enabled_tools to block specific tools.

Feature Flags

Explicit opt-in/opt-out toggles for experimental features.

Multi-AgentEnable multi-agent tools (experimental).

Unified ExecUse PTY-backed exec tool.

UndoEnable undo support.

Shell ToolEnable the shell tool.

Personality FeatureEnable the personality system for supported models.

Fast ModeEnable fast mode for quicker responses.

SQLiteEnable SQLite-backed state storage.

Image GenerationEnable image generation capabilities.

Agents

Multi-agent roles and concurrency limits.

Max Threads

Maximum concurrent agent threads.

Max Depth

Maximum nesting depth for agent spawning.

Job Max Runtime (sec)

Maximum runtime in seconds for a worker agent job.

OpenTelemetry

Configure OpenTelemetry log and trace export.

Log User PromptInclude user prompts in OpenTelemetry logs.

Environment

Environment tag for OpenTelemetry exports.

Log Exporter

OpenTelemetry log exporter target.

Trace Exporter

OpenTelemetry trace exporter target.

Windows

Windows-specific sandbox and compatibility settings.

Windows Sandbox

Sandbox mode for Windows environments.

UnelevatedRun without elevated privileges.

WSL Setup AcknowledgedSuppress the WSL setup prompt on Windows.