Developer integration
Codex CLI
Configure Codex to use Albatross as a custom Responses provider.
- Protocol
- OpenAI Responses
- Support status
- Pending revalidation · last tested with Codex CLI 0.153.4
Configure the provider
Put the provider definition in the user-level ~/.codex/config.toml. Keep the API key in an environment variable; env_key is the variable name, not the key itself. Back up and merge the provider settings into the existing file. Persist the variable in the startup file for the OS and shell that launches Codex; the file-content example below is only for macOS zsh. Save both files, open a new terminal, enter your practice folder and run the separate launch command. Restart an existing Codex process to load the changes. Never paste the launch command into ~/.zshrc. On Windows use your shell's environment-variable syntax instead of export.
model = "<ALBATROSS_MODEL>"
model_provider = "albatross"
[model_providers.albatross]
name = "Albatross"
base_url = "https://staging-gateway.bulakua.com/v1"
env_key = "ALBATROSS_API_KEY"
wire_api = "responses"
requires_openai_auth = false
request_max_retries = 0
stream_max_retries = 0export ALBATROSS_API_KEY="<ALBATROSS_API_KEY>"codexFirst acceptance
- Ask for a short text response.
- Ask Codex to read a non-sensitive test file.
- Allow a harmless command in a disposable directory.
- Ask it to modify a temporary test file and inspect the diff.
- Cancel one in-progress request and confirm it is not shown as complete.
- Use an invalid synthetic key and confirm the 401 is understandable.
Important boundaries
- Codex runs tools locally; Albatross does not execute them.
- A successful CLI test does not prove Codex Desktop compatibility.
- Custom-provider Responses calls are stateless and require opaque response state to survive tool continuation.
- Repeat acceptance after changing Codex, Gateway, model alias, or provider route versions.
Switch models with one provider
Keep one Albatross provider, base URL and environment key. Set a default model, override it for one run with --model, or use /model in the interactive UI.
- Custom aliases may require model_catalog_json. Include every selectable alias and restart Codex after changing the catalog.
- Use aliases returned by GET /v1/models for this key.
codex --model '<ALBATROSS_TEXT_QUALITY_ALIAS>'
codex exec -m '<ALBATROSS_TEXT_FAST_ALIAS>' 'Inspect the test failures'