Optionally fetch list of models from API

[keegancsmith]

A lot of endpoints support getting a list of the supported models. Right now gptel hardcodes the list of supported models, or we need to configure it via :models when we create the backend. It would be cool if it was possible to make this dynamic. EG maybe the models slot could also be a function?

[gregoryg]

@keegancsmith I get a list of models only for my self-hosted Ollama service.

(setq gjg/ollama-api-endpoint "http://1.2.3.4:11434")
(defun gjg/get-ollama-models ()
  "Return a list of Ollama models from the API, or nil if the API is not available."
  (when-let*
      ((url (url-generic-parse-url gjg/ollama-api-endpoint))
       (host (url-host url))
       (port (url-port url))
       ;; Use short timeout with curl to assure API port is open
       (maybe-ollama-models (s-split "\n"
                                     (shell-command-to-string (concat "curl -s --connect-timeout 0.5 '"
                                                                      gjg/ollama-api-endpoint "/api/tags' | jq -r '.models[].name'"))))
       (my-ollama-models (seq-filter (lambda (s) (not (string= s "")))
                                     maybe-ollama-models)))
    (gptel-make-ollama
     "Ollama"
     :host (concat host ":" (number-to-string port))
     :models my-ollama-models
     :stream nil)
    my-ollama-models))

;; set globally, as ollama-models is relevant to more than gptel
(setq ollama-models (gjg/get-ollama-models))

;; if API port was not found, remove backend
(unless ollama-models
  (setq gptel--known-backends
        (assoc-delete-all "Ollama" gptel--known-backends #'equal)))