Docs/Coding agents

Codex CLI

Run OpenAI's Codex CLI on your Osarai instance through a custom model provider that speaks the Responses API.

Codex CLI is OpenAI's coding agent for the terminal. It talks to custom providers over the Responses API only, which Osarai serves at https://osarai.xyz/v1/responses.

#Install

npm install -g @openai/codex

brew install --cask codex

#Configure

Add a provider to Codex's configuration file and make it the default:

model = "gpt-oss-20b"
model_provider = "osarai"
model_context_window = 131072
model_auto_compact_token_limit = 110000
model_reasoning_effort = "medium"

[model_providers.osarai]
name = "Osarai"
base_url = "https://osarai.xyz/v1"
env_key = "OSARAI_API_KEY"
wire_api = "responses"

model = "gpt-oss-20b"
model_provider = "osarai"
model_context_window = 131072
model_auto_compact_token_limit = 110000
model_reasoning_effort = "medium"

[model_providers.osarai]
name = "Osarai"
base_url = "https://osarai.xyz/v1"
env_key = "OSARAI_API_KEY"
wire_api = "responses"

#Run

export OSARAI_API_KEY="qck_your_key_here"
cd your-project
codex

The header shows model: gpt-oss-20b and provider: osarai. To run one prompt and exit:

codex exec "Explain what src/math.ts does" < /dev/null

codex exec reads extra input from standard input when it is not a terminal; < /dev/null stops it waiting in scripts and CI.

#Switching models

A profile is a second file next to config.toml, named <profile>.config.toml. Its keys override the main file, so the provider carries over:

model = "qwen3.8-27b"
model_context_window = 131072

codex --profile qwen

#Troubleshooting