> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/docsfab/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/docsfab/_mcp/server. # Configure the Model in DocsFab > Change the DocsFab provider, model, API, or endpoint interactively or with setup flags. Run `docsfab setup` again to change the model configuration that DocsFab uses. This procedure configures a provider directly. To use a recorded model and endpoint instead, follow [Select an Installed Composition](/docsfab/configure/installed-compositions). ## Prerequisites - Install the CLI by following [Install DocsFab](/docsfab/get-started/installation). - Obtain the provider's model ID and API key. Direct OpenAI and Anthropic selections use keys from their product consoles. Compatible selections use the hosting service's key and base endpoint URL, including models hosted on AWS, Azure, or another service. - Confirm that the endpoint supports streaming, native tool calling, and the API you select. An OpenAI Chat Completions server does not necessarily support OpenAI Responses. Refer to [Set Up DocsFab](/docsfab/get-started/installation#set-up-docsfab) for the four provider choices, and to [Provider Selection Defaults](/docsfab/reference/status-and-configuration/model-profile-fields#provider-selection-defaults) for the default APIs and credential variable names. DocsFab has no default model. ## Steps 1. Run `docsfab setup` in a terminal, or supply the provider and model as flags. Use one of the following commands, replacing `MODEL_ID` and the example URL with values from your provider: ```bash docsfab setup --provider openai --model MODEL_ID docsfab setup --provider openai-compatible --model MODEL_ID \ --endpoint https://models.example.com/v1 docsfab setup --provider anthropic --model MODEL_ID docsfab setup --provider anthropic-compatible --model MODEL_ID \ --endpoint https://models.example.com ``` 2. For either OpenAI option, select the API that the endpoint supports. Interactive setup offers **Responses** and **Chat Completions**. To supply the choice as a flag, use `--api openai-responses` or `--api openai-chat-completions`. Both Anthropic options use Messages. 3. Enter the required key at the hidden **Model API key** prompt. Setup rejects blank input and stores the key in the owner-private `secrets.env` file. Interactive setup asks for the key each time, even if one is already saved. When you keep the same provider, applicable prompts offer the saved model, endpoint, and API as defaults. Never put a key in a command-line argument. Refer to [Manage Local Configuration and Credentials](/docsfab/manage/local-configuration) for the storage location and permissions. ## Choose an Endpoint Compatible options require an endpoint that implements the selected API and accepts an API key. They do not configure native AWS or Azure SDK protocols or cloud-specific authentication. Provide a base URL, not the full request path. OpenAI protocols append `/chat/completions` or `/responses`. If you paste the full request URL for the selected API, setup removes that path so it is not sent twice. An endpoint saved with a request path before this behavior existed keeps failing until you rerun `docsfab setup` and accept the shown endpoint. Anthropic accepts a base URL with or without a trailing `/v1` and sends requests to `/v1/messages` without duplicating the version segment. Its Pi adapter adds the exact `?beta=true` query required by that SDK version. Gateway prefixes are preserved. Setup shows the final `Request URL` for your selected API. Supplied base URLs cannot contain credentials, query strings, or fragments. An Anthropic-compatible gateway must accept the displayed beta query. Endpoint URLs use HTTPS. The one exception is a model server on the same machine: DocsFab accepts plain `http://` only when the host is `localhost`, an IPv4 address in `127.0.0.0/8` such as `127.0.0.1`, or `[::1]`. Any other plain HTTP host, including a name such as `localhost.example.com`, stops with envelope code `model-endpoint-insecure-transport` before setup writes anything. A hand-edited `MODEL_BASE_URL` with such a host stops with the same code when DocsFab loads the configuration. For a local model server, refer to [Use a Local NIM](/docsfab/configure/local-nim). ## Set Up Without a Terminal Without an attached terminal, supply the choices as flags and provide the corresponding credential variable through your host's secret facility. `--json` also disables prompts. Without a terminal, `docsfab setup` with no `--provider`, `--profile`, `--preset`, or `--migrate` option stops with envelope code `setup-model-choice-required` before it writes anything. Its `Next action` lists the choices: run setup in a terminal, pass `--provider` and `--model`, or pass `--profile` with a composition id. A stored or exported API key does not choose a model. A host that already has a live model selection keeps it. If a saved selection stops with `composition-retired`, follow [Model Selection Is No Longer Supported](/docsfab/troubleshooting#model-selection-is-no-longer-supported). An explicit provider selection requires its key in the environment for that invocation; setup does not reuse a saved key when selecting another endpoint. Never put a key in a command-line argument. ## Check the Result A successful setup summary reports `DocsFab setup complete` and `Status: Ready`. It also shows the selected model, provider, API, endpoint, final request URL, effective context limit, configuration location, and next command. The `Connection check: Not run` line confirms that setup validated only the local configuration. In a terminal, setup highlights the selected option and groups the summary into model, configuration, and next-step sections. Set `NO_COLOR=1` to disable colors. Redirected output remains plain text, and `--json` returns structured data. To inspect the saved selection, run `docsfab profile inspect`. To test the connection, use `docsfab doctor --live`, as described in [Check DocsFab Readiness](/docsfab/manage/check-readiness). ## Related Topics - [Select an Installed Composition](/docsfab/configure/installed-compositions) - [Model Profile Field Reference](/docsfab/reference/status-and-configuration/model-profile-fields) - [Model Calls Fail or the Model Key Is Rejected](/docsfab/troubleshooting#model-calls-fail-or-the-model-key-is-rejected)