Skip to main content
Codex CLI lets you add any OpenAI-compatible provider. Point it at Deepshi and you can code with Deepshi’s models from your terminal.

​
Configure the provider

Add Deepshi to ~/.codex/config.toml: 
# Make Deepshi the default
model_provider = "deepshi"
model = "deepshi-3.0"

[model_providers.deepshi]
name = "Deepshi"
base_url = "https://api.deepshi.ai/v1"
env_key = "DEEPSHI_API_KEY"
wire_api = "responses"
The top-level model and model_provider keys must come before the [model_providers.deepshi] block. In TOML, any keys written after a table header belong to that table. Set your key in the environment: 
export DEEPSHI_API_KEY="YOUR_DEEPSHI_API_KEY"

​
Run it

codex
Or select the provider and model inline for a single run: 
codex --config model_provider=deepshi --model deepshi-3.0
Use a model with the tools capability for agentic coding, since Codex relies on tool calling. See Text models.
wire_api = "responses" tells Codex to use the OpenAI Responses API. Current Codex versions speak only the Responses API, which Deepshi supports at https://api.deepshi.ai/v1. See Authentication for keys.