New Keys Per Release

• rust-v0.29.0 (2025-09-03)

  • [tools].view_image — allow attaching local images to conversations.
  • disable_paste_burst — disable multi-line paste burst submissions in TUI.
  • use_experimental_reasoning_summary — enable experimental reasoning summaries in UI.

• rust-v0.25.0 (2025-08-27)

  • No new config keys compared to v0.24.0. This release focused on TUI polish, bug fixes, and minor exec/stream behavior.

• rust-v0.24.0 (2025-08)

  • preferred_auth_method = "apikey" | "chatgpt" — choose ChatGPT login or API key when both are available.
  • [projects."/abs/path"].trust_level = "trusted" — mark specific repos as trusted to reduce prompts.
  • model_verbosity = "low" | "medium" | "high" — controls text verbosity for GPT‑5 models.
  • [model_providers.].request_max_retries — per‑provider HTTP retry cap.
  • [model_providers.].stream_max_retries — per‑provider stream reconnect cap.
  • [model_providers.].stream_idle_timeout_ms — per‑provider stream idle timeout.
  • show_raw_agent_reasoning = true — show raw reasoning content when available.
  • [shell_environment_policy].experimental_use_profile = true — run commands through a profile shell (advanced).
  • chatgpt_base_url = "https://chatgpt.com/backend-api/" — override ChatGPT backend URL (advanced).
  • responses_originator_header_internal_override = "codex_cli_rs" — internal/testing only.
  • [tools].web_search = true (alias: web_search_request) — enable native web search tool.

Codex_Utils

Quick Start

1. Copy config.toml to ~/.codex/config.toml
2. Authenticate (choose one):
   • ChatGPT sign‑in: run codex and choose “Sign in with ChatGPT” (no API key needed).
   • API key: set OPENAI_API_KEY if you prefer usage‑based billing.
     • macOS/Linux: export OPENAI_API_KEY="your-key"
     • Windows (PowerShell): $env:OPENAI_API_KEY="your-key"
3. Run: codex

Key Configuration Options

Core Settings

model = "gpt-5"                    # or "o3", "o4-mini"
model_reasoning_effort = "high"    # minimal | low | medium | high
approval_policy = "never"          # untrusted | on-failure | on-request | never
sandbox_mode = "danger-full-access" # read-only | workspace-write | danger-full-access

# Network (applies when sandbox_mode = "workspace-write")
[sandbox_workspace_write]
network_access = false             # allow outbound network from sandbox

# UI/UX
file_opener = "vscode"             # Opens files in VS Code
show_raw_agent_reasoning = true    # Show raw reasoning content (if available)
notify = ["notify-send", "Codex"]  # Desktop notifications
use_experimental_reasoning_summary = false  # Experimental reasoning summaries
disable_paste_burst = false                # Prevent rapid multi-line paste submissions

# Tools
[tools]
web_search = true                  # Enable web search

view_image = true                    # Attach local images as context
# History
[history]
persistence = "save-all"           # Keep conversation history

Security: Shell Environment Policy

[shell_environment_policy]
ignore_default_excludes = false   # Blocks KEY/SECRET/TOKEN vars
include_only = [
    # Core essentials
    "PATH", "HOME", "USER", "SHELL",
    "LANG", "LC_*", "TERM", "EDITOR",
    "PWD", "TMPDIR", "XDG_*",
    
    # Development (add as needed)
    "PYTHON*", "CONDA_*", "VIRTUAL_ENV*",  # Python
    "NODE_*", "NPM_*", "NVM_*",            # Node.js
]
# Never include: "*_KEY", "*_SECRET", "*_TOKEN", "AWS_*", "AZURE_*"

Trust Settings

[projects."/path/to/your/repo"]
trust_level = "trusted"            # Skip security prompts for specific repos

Providers

OpenAI (Default)

• Set: OPENAI_API_KEY
• Optional: OPENAI_ORGANIZATION, OPENAI_PROJECT

### Ollama (Local)
```toml
[model_providers.ollama]
base_url = "http://localhost:11434/v1"

Profiles (

Switch configurations with codex --profile <name>:

[profiles.quick]
model_reasoning_effort = "minimal"  # Fast mode

[profiles.secure]
approval_policy = "on-request"      # Careful mode
sandbox_mode = "workspace-write"    # Restricted

Files in This Repo

• config.toml - Production-ready configuration with security defaults
• codex-config-comprehensive.toml - Full reference with all available options

Tips

• Start with the provided config.toml and customize as needed
• Use include_only in shell_environment_policy for maximum security
• Enable show_raw_agent_reasoning to understand the assistant's decisions
• Set up project trust to reduce security prompts in your repos

Resources

• Official Codex Documentation
• Environment Variables Security