How to use Ollama models with Codex

Using Ollama models with Codex keeps prompt traffic on the local machine while still letting Codex inspect files, edit code, and run commands in the current workspace.

The Codex CLI routes local-model runs through --oss, and --local-provider ollama keeps the request on Ollama's built-in OpenAI-compatible API at http://localhost:11434/v1. The value passed with -m must match the exact model ID exposed by Ollama.

The Ollama server must already be running and the target model must already be available before the prompt run starts. Running the command from a Git repository, or pointing Codex at one with -C, keeps the trusted-directory check in place so the prompt reaches Ollama instead of stopping at workspace validation.

1. Check Ollama for the exact model ID that Codex should request.

   $ ollama list
   NAME           ID              SIZE     MODIFIED
   gpt-oss:20b    17052f91a42e    13 GB    4 months ago

   Use the value in the NAME column as the -m model argument, and pull the model first if it does not already appear in this list.

2. Run Codex against the Ollama provider and point it at the repository directory that Codex should inspect.

   $ codex exec --oss \
   --local-provider ollama -m gpt-oss:20b \
   -C ~/repo "Reply with exactly: OK"
   OK

   --local-provider ollama avoids accidental routing to another local backend, and -C prevents a trusted-directory failure when the current shell is outside the target repo.

   Related: How to fix Codex trusted directory error

3. Save the final Ollama reply to a file when the result needs reuse outside the terminal.

   $ codex exec --oss \
   --local-provider ollama -m gpt-oss:20b \
   -C ~/repo -o /tmp/ok.txt \
   "Reply with exactly: OK"
   OK

   -o writes only the final assistant message and overwrites the destination file when it already exists.

   Related: How to save the last Codex response to a file

4. Confirm that the saved file contains the expected local-model response.

   $ cat /tmp/ok.txt
   OK