Improve ollama model listing getting capabilities, avoiding change ollama config for different models.

Author: ericdallo

CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## Unreleased
 
+- Improve ollama model listing getting capabilities, avoiding change ollama config for different models.
+
 ## 0.13.0
 
 - Support reasoning for ollama models that support think.

docs/configuration.md

@@ -193,7 +193,7 @@ interface Config {
   "ollama" : {
     "host" : "http://localhost",
     "port" : 11434,
-    "useTools": false,
+    "useTools": true,
     "think": true
   },
   "chat" : {

src/eca/db.clj

@@ -13,20 +13,24 @@
    :chat-default-behavior "agent"
    :models {"o4-mini" {:tools true
                        :web-search false
+                       :reason? true
                        :input-token-cost (/ 1.10 one-million)
                        :output-token-cost (/ 4.40 one-million)}
             "o3" {:tools true
                   :web-search false
+                  :reason? true
                   :input-token-cost (/ 2.0 one-million)
                   :output-token-cost (/ 8.0 one-million)}
             "gpt-4.1" {:tools true
                        :web-search true
+                       :reason? false
                        :max-output-tokens 32000
                        :input-token-cost (/ 2.0 one-million)
                        :output-token-cost (/ 8.0 one-million)}
             "claude-sonnet-4-0" {:tools true
                                  :web-search true
                                  :max-output-tokens 8196
+                                 :reason? true
                                  :reason-tokens 2048
                                  :input-token-cost (/ 3.0 one-million)
                                  :input-cache-creation-token-cost (/ 3.75 one-million)
@@ -35,13 +39,15 @@
             "claude-opus-4-0" {:tools true
                                :web-search true
                                :max-output-tokens 8196
+                               :reason? true
                                :reason-tokens 2048
                                :input-token-cost (/ 15.0 one-million)
                                :input-cache-creation-token-cost (/ 18.75 one-million)
                                :input-cache-read-token-cost (/ 1.5 one-million)
                                :output-token-cost (/ 75.0 one-million)}
             "claude-3-5-haiku-latest" {:tools true
                                        :web-search true
+                                       :reason? false
                                        :max-output-tokens 4096
                                        :input-token-cost (/ 0.8 one-million)
                                        :input-cache-creation-token-cost (/ 1.0 one-million)

src/eca/handlers.clj

@@ -28,15 +28,14 @@
             {}
             custom-providers)))
   (when-let [ollama-models (seq (llm-api/extra-models config))]
-    (swap! db* update :models merge
-           (reduce
-            (fn [models {:keys [model]}]
-              (assoc models
-                     (str config/ollama-model-prefix model)
-                     {:tools (get-in config [:ollama :useTools] false)
-                      :think (get-in config [:ollama :think] false)}))
-            {}
-            ollama-models))))
+    (let [models (reduce
+                  (fn [models {:keys [model] :as ollama-model}]
+                    (assoc models
+                           (str config/ollama-model-prefix model)
+                           (select-keys ollama-model [:tools :reason?])))
+                  {}
+                  ollama-models)]
+      (swap! db* update :models merge models))))
 
 (defn initialize [{:keys [db* config]} params]
   (logger/logging-task
@@ -81,8 +80,8 @@
 
 (defn chat-tool-call-approve [{:keys [db*]} params]
   (logger/logging-task
-      :eca/chat-tool-call-approve
-      (f.chat/tool-call-approve params db*)))
+   :eca/chat-tool-call-approve
+   (f.chat/tool-call-approve params db*)))
 
 (defn chat-tool-call-reject [{:keys [db*]} params]
   (logger/logging-task

src/eca/llm_api.clj

@@ -12,8 +12,17 @@
 (def ^:private logger-tag "[LLM-API]")
 
 (defn extra-models [config]
-  (llm-providers.ollama/list-models {:host (:host (:ollama config))
-                                     :port (:port (:ollama config))}))
+  (let [ollama-host (:host (:ollama config))
+        ollama-port (:port (:ollama config))]
+    (mapv
+     (fn [{:keys [model] :as ollama-model}]
+       (let [capabilities (llm-providers.ollama/model-capabilities {:host ollama-host :port ollama-port :model model})]
+         (assoc ollama-model
+                :tools (and (get-in config [:ollama :useTools] true)
+                            (boolean (some #(= % "tools") capabilities)))
+                :reason? (and (get-in config [:ollama :think] true)
+                              (boolean (some #(= % "thinking") capabilities))))))
+     (llm-providers.ollama/list-models {:host ollama-host :port ollama-port}))))
 
 ;; TODO ask LLM for the most relevant parts of the path
 (defn refine-file-context [path lines-range]
@@ -116,7 +125,7 @@
         :user-messages user-messages
         :max-output-tokens max-output-tokens
         :reason-tokens reason-tokens
-        :reason? reason?
+        :reason? (and reason? (:reason? model-config))
         :past-messages past-messages
         :tools tools
         :web-search web-search
@@ -133,7 +142,7 @@
         :user-messages user-messages
         :max-output-tokens max-output-tokens
         :reason-tokens reason-tokens
-        :reason? reason?
+        :reason? (and reason? (:reason? model-config))
         :past-messages past-messages
         :tools tools
         :web-search web-search
@@ -145,7 +154,7 @@
       (llm-providers.ollama/completion!
        {:host (-> config :ollama :host)
         :port (-> config :ollama :port)
-        :reason? (and reason? (:think model-config))
+        :reason? (and reason? (:reason? model-config))
         :model (string/replace-first model config/ollama-model-prefix "")
         :instructions instructions
         :user-messages user-messages
@@ -168,7 +177,7 @@
           :user-messages user-messages
           :max-output-tokens max-output-tokens
           :reason-tokens reason-tokens
-          :reason? reason?
+          :reason? (and reason? (:reason? model-config))
           :past-messages past-messages
           :web-search web-search
           :tools tools

src/eca/llm_providers/ollama.clj

@@ -12,6 +12,7 @@
 
 (def ^:private chat-url "%s/api/chat")
 (def ^:private list-models-url "%s/api/tags")
+(def ^:private show-model-url "%s/api/show")
 
 (defn ^:private base-url [host port]
   (or (System/getenv "OLLAMA_API_BASE")
@@ -35,6 +36,25 @@
       (logger/warn logger-tag "Error listing running models:" (ex-message e))
       [])))
 
+(defn model-capabilities [{:keys [host port model]}]
+  (try
+    (let [rid (llm-util/gen-rid)
+          {:keys [status body]} (http/post
+                                 (format show-model-url (base-url host port))
+                                 {:throw-exceptions? false
+                                  :body (json/generate-string {:model model})
+                                  :as :json})]
+      (if (= 200 status)
+        (do
+          (llm-util/log-response logger-tag rid "api/show" body)
+          (:capabilities body))
+        (do
+          (logger/warn logger-tag "Unknown status code:" status)
+          [])))
+    (catch Exception e
+      (logger/warn logger-tag "Error getting model:" (ex-message e))
+      [])))
+
 (defn ^:private base-completion-request! [{:keys [rid url body on-error on-response]}]
   (llm-util/log-request logger-tag rid url body)
   (let [reason-id (str (random-uuid))