Configuring Codex

Codex should work out of the box for most users. But sometimes you want to configure Codex to your own liking to better suit your needs. For this there is a wide range of configuration options.

Codex configuration file

The configuration file for Codex is located at ~/.codex/config.toml.

To access the configuration file when you are using the Codex IDE extension, you can click the gear icon in the top right corner of the extension and then clicking Codex Settings > Open config.toml.

This configuration file is shared between the CLI and the IDE extension and can be used to configure things like the default model, approval policies, sandbox settings or MCP servers that Codex should have access to.

High level configuration options

Codex provides a wide range of configuration options. Some of the most commonly changed settings are:

Default model

Pick which model Codex uses by default in both the CLI and IDE.

Using config.toml:

model = "gpt-5"

Using CLI arguments:

codex --model gpt-5

Model provider

Select the backend provider referenced by the active model. Be sure to define the provider in your config first.

Using config.toml:

model_provider = "ollama"

Using CLI arguments:

codex --config model_provider="ollama"

Approval prompts

Control when Codex pauses to ask before running generated commands.

Using config.toml:

approval_policy = "on-request"

Using CLI arguments:

codex --ask-for-approval on-request

Sandbox level

Adjust how much filesystem and network access Codex has while executing commands.

Using config.toml:

sandbox_mode = "workspace-write"

Using CLI arguments:

codex --sandbox workspace-write

Reasoning depth

Tune how much reasoning effort the model applies when supported.

Using config.toml:

model_reasoning_effort = "high"

Using CLI arguments:

codex --config model_reasoning_effort="high"

Command environment

Restrict or expand which environment variables are forwarded to spawned commands.

Using config.toml:

[shell_environment_policy]
include_only = ["PATH", "HOME"]

Using CLI arguments:

codex --config shell_environment_policy.include_only='["PATH","HOME"]'

Profiles

Profiles bundle a set of configuration values so you can jump between setups without editing config.toml each time. They currently apply to the Codex CLI.

Define profiles under [profiles.<name>] in config.toml and launch the CLI with codex --profile <name>:

model = "gpt-5-codex"
approval_policy = "on-request"

[profiles.deep-review]
model = "gpt-5-pro"
model_reasoning_effort = "high"
approval_policy = "never"

[profiles.lightweight]
model = "gpt-4.1"
approval_policy = "untrusted"

Running codex --profile deep-review will use the gpt-5-pro model with high reasoning effort and no approval policy. Running codex --profile lightweight will use the gpt-4.1 model with untrusted approval policy. To make one profile the default, add profile = "deep-review" at the top level of config.toml; the CLI will load that profile unless you override it on the command line.

Values resolve in this order: explicit CLI flags (like --model) override everything, profile values come next, then root-level entries in config.toml, and finally the CLI’s built-in defaults. Use that precedence to layer common settings at the top level while letting each profile tweak just the fields that need to change.

Feature flags

Optional and experimental capabilities are toggled via the [features] table in config.toml. If Codex emits a deprecation warning mentioning a legacy key (such as experimental_use_exec_command_tool), move that setting into [features] or launch the CLI with codex --enable <feature>.

[features]
streamable_shell = true          # enable the streamable exec tool
web_search_request = true        # allow the model to request web searches
# view_image_tool defaults to true; omit to keep defaults

Supported features

Enabling features quickly

• In config.toml: add feature_name = true under [features].
• CLI onetime: codex --enable feature_name.
• Multiple flags: codex --enable feature_a --enable feature_b.
• Disable explicitly by setting the key to false in config.toml.

Advanced configuration

Custom model providers

Define additional providers and point model_provider at them:

model = "gpt-4o"
model_provider = "openai-chat-completions"

[model_providers.openai-chat-completions]
name = "OpenAI using Chat Completions"
base_url = "https://api.openai.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"
query_params = {}

[model_providers.ollama]
name = "Ollama"
base_url = "http://localhost:11434/v1"

[model_providers.mistral]
name = "Mistral"
base_url = "https://api.mistral.ai/v1"
env_key = "MISTRAL_API_KEY"

Add request headers when needed:

[model_providers.example]
http_headers = { "X-Example-Header" = "example-value" }
env_http_headers = { "X-Example-Features" = "EXAMPLE_FEATURES" }

Azure provider & per-provider tuning

[model_providers.azure]
name = "Azure"
base_url = "https://YOUR_PROJECT_NAME.openai.azure.com/openai"
env_key = "AZURE_OPENAI_API_KEY"
query_params = { api-version = "2025-04-01-preview" }
wire_api = "responses"

[model_providers.openai]
request_max_retries = 4
stream_max_retries = 10
stream_idle_timeout_ms = 300000

Model reasoning, verbosity, and limits

model_reasoning_summary = "none"          # disable summaries
model_verbosity = "low"                   # shorten responses on Responses API providers
model_supports_reasoning_summaries = true # force reasoning on custom providers
model_context_window = 128000             # override when Codex doesn't know the window
model_max_output_tokens = 4096            # cap completion length

model_verbosity applies only to providers using the Responses API; Chat Completions providers will ignore the setting.

Approval policies and sandbox modes

Pick approval strictness (affects when Codex pauses) and sandbox level (affects file/network access). See Sandbox & approvals for deeper examples.

approval_policy = "untrusted"   # other options: on-request, on-failure, never
sandbox_mode = "workspace-write"

[sandbox_workspace_write]
exclude_tmpdir_env_var = false  # allow $TMPDIR
exclude_slash_tmp = false       # allow /tmp
writable_roots = ["/Users/YOU/.pyenv/shims"]
network_access = false          # opt in to outbound network

Disable sandboxing entirely (use only if your environment already isolates processes):

sandbox_mode = "danger-full-access"

Shell environment templates

shell_environment_policy controls which environment variables Codex passes to any subprocess it launches (for example, when running a tool-command the model proposes). Start from a clean slate (inherit = "none") or a trimmed set (inherit = "core"), then layer on excludes, includes, and overrides to avoid leaking secrets while still providing the paths, keys, or flags your tasks need.

[shell_environment_policy]
inherit = "none"
set = { PATH = "/usr/bin", MY_FLAG = "1" }
ignore_default_excludes = false
exclude = ["AWS_*", "AZURE_*"]
include_only = ["PATH", "HOME"]

Patterns are case-insensitive globs (*, ?, [A-Z]); ignore_default_excludes = false keeps the automatic KEY/SECRET/TOKEN filter before your includes/excludes run.

MCP servers

See the dedicated MCP guide for full server setups and toggle descriptions. Below is a minimal STDIO example using the Context7 MCP server:

[mcp_servers.context7]
command = "npx"
args = ["-y", "@upstash/context7-mcp"]

Observibility and telemetry

Enable OpenTelemetry (Otel) log export to track Codex runs (API requests, SSE/events, prompts, tool approvals/results). Disabled by default; opt in via [otel]:

[otel]
environment = "staging"   # defaults to "dev"
exporter = "none"         # set to otlp-http or otlp-grpc to send events
log_user_prompt = false   # redact user prompts unless explicitly enabled

Choose an exporter:

[otel]
exporter = { otlp-http = {
  endpoint = "https://otel.example.com/v1/logs",
  protocol = "binary",
  headers = { "x-otlp-api-key" = "${OTLP_TOKEN}" }
}}

[otel]
exporter = { otlp-grpc = {
  endpoint = "https://otel.example.com:4317",
  headers = { "x-otlp-meta" = "abc123" }
}}

If exporter = "none" Codex records events but sends nothing. Exporters batch asynchronously and flush on shutdown. Event metadata includes service name, CLI version, env tag, conversation id, model, sandbox/approval settings, and per-event fields (see Config reference table below).

Notifications

Use notify to trigger an external program whenever Codex emits supported events (today: agent-turn-complete). This is handy for desktop toasts, chat webhooks, CI updates, or any side-channel alerting that the built-in TUI notifications don’t cover.

notify = ["python3", "/path/to/notify.py"]

Example notify.py (truncated) that reacts to agent-turn-complete:

#!/usr/bin/env python3
import json, subprocess, sys

def main() -> int:
    notification = json.loads(sys.argv[1])
    if notification.get("type") != "agent-turn-complete":
        return 0
    title = f"Codex: {notification.get('last-assistant-message', 'Turn Complete!')}"
    message = " ".join(notification.get("input-messages", []))
    subprocess.check_output([
        "terminal-notifier",
        "-title", title,
        "-message", message,
        "-group", "codex-" + notification.get("thread-id", ""),
        "-activate", "com.googlecode.iterm2",
    ])
    return 0

if __name__ == "__main__":
    sys.exit(main())

Place the script somewhere on disk and point notify to it. For lighter in-terminal alerts, toggle tui.notifications instead.

Personalizing the Codex IDE Extension

Additionally to configuring the underlying Codex agent through your config.toml file, you can also configure the way you use the Codex IDE extension.

To see the list of available configuration options, click the gear icon in the top right corner of the extension and then click IDE settings.

To define your own keyboard shortcuts to trigger Codex or add something to the Codex context, you can click the gear icon in the top right corner of the extension and then click Keyboard shortcuts.