Agentsflare Codex Configuration Guide

Codex is OpenAI's official CLI coding tool that supports natural language instructions for code writing, debugging, and refactoring.

Installing Codex

Install globally via npm:

npm install -g @openai/codex

Or run directly with npx (no installation required):

npx @openai/codex

Verify installation:

codex --version

Configuring Codex with Agentsflare

Codex uses a TOML configuration file, not YAML. See the official configuration basics and configuration reference. To ensure requests use Agentsflare instead of Codex's built-in OpenAI provider, define an explicit custom provider and select it as the default provider.

1. Create the user-level configuration file

Create or modify the following file:

• macOS / Linux: ~/.codex/config.toml
• Windows: %USERPROFILE%\\.codex\\config.toml

Put the following content in config.toml:

model = "gpt-5.3-codex"
model_provider = "agentsflare"

[model_providers.agentsflare]
name = "Agentsflare"
base_url = "https://api.agentsflare.com/v1"
env_key = "AGENTSFLARE_API_KEY"
requires_openai_auth = false
wire_api = "responses"

2. Provide the Agentsflare API key

Set the environment variable before launching Codex:

export AGENTSFLARE_API_KEY="Your Agentsflare API Key"
codex

On Windows PowerShell:

$env:AGENTSFLARE_API_KEY = "Your Agentsflare API Key"
codex

env_key tells Codex which environment variable supplies the key. Keep the key out of config.toml and do not use OPENAI_API_KEY as the only configuration: that variable is associated with Codex's built-in OpenAI provider and does not select the Agentsflare provider.

Endpoint Rules

This configuration uses the Responses API, which is the protocol Codex expects for this provider:

Item	Value
Provider base URL in config.toml	https://api.agentsflare.com/v1
Actual request sent by Codex	POST https://api.agentsflare.com/v1/responses
Required wire_api	"responses"

Do not put /responses or /chat/completions in base_url. Codex appends the route selected by wire_api; including a full endpoint path can produce an invalid duplicated path. Do not configure Codex with wire_api = "chat" for this integration.

Why the Previous Configuration Did Not Work

The previous example could leave Codex on the OpenAI address for several reasons:

1. ~/.codex/config.yaml is not Codex's user configuration file. The file is ~/.codex/config.toml.
2. Top-level api_key and base_url do not define a selected custom provider. The provider must be declared under [model_providers.<id>], and model_provider must point to that ID.
3. OPENAI_BASE_URL and OPENAI_API_KEY configure the built-in OpenAI path; they do not replace the explicit custom-provider selection shown above.
4. A project-local .codex/config.toml cannot override machine-local provider settings. Put model_provider and [model_providers.agentsflare] in the user-level ~/.codex/config.toml.

Verify the Active Route

Run the diagnostic first:

codex doctor

The report should show all of the following:

• config.toml loaded from ~/.codex/config.toml;
• default model provider agentsflare;
• OpenAI authentication not required;
• custom API base URL https://api.agentsflare.com/v1;
• a reachable provider route.

Then send one real request:

codex exec --skip-git-repo-check "Reply with exactly: AGENTSFLARE_OK"

If the request still reaches an OpenAI address, check that model_provider = "agentsflare" is at the top level of the user config, that the section is named exactly [model_providers.agentsflare], and that the command is not being run with --ignore-user-config or a different CODEX_HOME.

Common Commands

Start an interactive session:

codex

Execute a single instruction directly:

codex "Refactor utils.js into TypeScript"

Launch with a specific model:

codex --model gpt-5.3-codex

View help:

codex --help