Hooking Codex CLI up to skiapi.dev to run the free DeepSeek V3.2【1】
Hooking Codex CLI up to skiapi.dev to run code with the free DeepSeek V3.2
This is episode 1 of hooking OpenAI Codex CLI up to skiapi.dev to run free models. skiapi.dev is a third-party API aggregator that serves the free DeepSeek V3.2 and Qwen-3-coder-next. The idea is to use the token it hands you to stand in for the official OpenAI key, so Codex thinks it’s still talking to OpenAI.
Setup is two steps. First, set environment variables in PowerShell: put the token in OPENAI_API_KEY, point OPENAI_BASE_URL at https://skiapi.dev, then use SetEnvironmentVariable to persist them at the User level.
Second, edit %USERPROFILE%.codex\config.toml and add a [model_providers.skiapi] block: point base_url at https://skiapi.dev/v1, set wire_api to “responses”, and up top set model_provider = “skiapi”, model to gpt-5.4, and crank reasoning_effort to xhigh. Easy gotcha: the env var base_url has no /v1, while the config one does have /v1 — don’t mix them up. wire_api = “responses” is the key: Codex only speaks OpenAI’s responses protocol, so it’ll only connect once the proxy is aligned with it.
One heads-up: your token and requests all pass through skiapi.dev, a third party. Free is free, but don’t send sensitive code through it. The title is tagged 【1】, so there should be more parts coming.