Add usage tokens + cost to chat messages

Author: ericdallo

CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## Unreleased
 
+- Add usage tokens + cost to chat messages.
+
 ## 0.5.1
 
 - Fix openai key

docs/protocol.md

@@ -346,7 +346,7 @@ interface ChatPromptResponse {
      * The model used for this chat request.
      */
     model: ChatModel;
-
+    
     status: 'success';
 }
 ```
@@ -376,16 +376,6 @@ interface ChatContentReceivedParams {
      * The owner of this content.
      */
     role: 'user' | 'system' | 'assistant';
-
-    /**
-     * Optional metadata about the generation
-     */
-    metadata?: {
-        /**
-         * Number of tokens used in the generation
-         */
-        tokensUsed?: number;
-    };
 }
 
 /**
@@ -395,6 +385,7 @@ type ChatContent =
     | TextContent 
     | URLContent 
     | ProgressContent 
+    | UsageContent
     | FileChangeContent
     | ToolCallPrepareContent
     | ToolCallRunContent
@@ -429,20 +420,35 @@ interface URLContent {
 }
 
 /**
- * Details about the progress of the chat completion.
+ * Details about the chat's usage, like used tokens and costs.
  */
-interface ProgressContent {
-    type: 'progress';
-
+interface UsageContent {
+    type: 'usage';
+    
+    /*
+     * Number of tokens sent on previous prompt including all context used by ECA.
+     */
+    messageInputTokens: number;
+    
+    /*
+     * Number of tokens received from LLm in last prompt.
+     */
+    messageOutputTokens: number;
+    
+    /**
+     * The total input + output tokens of the whole chat session so far.
+     */
+    sessionTokens: number;
+    
     /**
-     * The statue of this progress
+     * The cost of the last sent message summing input + output tokens.
      */
-    state: 'running' | 'finished';
+    messageCost?: string; 
 
     /**
-     * The text detailing the progress
+     * The cost of the whole chat session so far.
      */
-    text?: string;
+    sessionCost?: string;
 }
 
 /**

src/eca/db.clj

@@ -2,6 +2,8 @@
 
 (set! *warn-on-reflection* true)
 
+(def ^:private one-million 1000000)
+
 (defonce initial-db
   {:client-info {}
    :workspace-folders []
@@ -10,17 +12,29 @@
    :chat-behaviors ["agent" "chat"]
    :chat-default-behavior "agent"
    :models {"o4-mini" {:tools true
-                       :web-search false}
+                       :web-search false
+                       :input-token-cost (/ 1.10 one-million)
+                       :output-token-cost (/ 4.40 one-million)}
             "o3" {:tools true
-                  :web-search false}
+                  :web-search false
+                  :input-token-cost (/ 2.0 one-million)
+                  :output-token-cost (/ 8.0 one-million)}
             "gpt-4.1" {:tools true
-                       :web-search true}
+                       :web-search true
+                       :input-token-cost (/ 2.0 one-million)
+                       :output-token-cost (/ 8.0 one-million)}
             "claude-sonnet-4-0" {:tools true
-                                 :web-search true}
+                                 :web-search true
+                                 :input-token-cost (/ 3.0 one-million)
+                                 :output-token-cost (/ 15.0 one-million)}
             "claude-opus-4-0" {:tools true
-                               :web-search true}
+                               :web-search true
+                               :input-token-cost (/ 15.0 one-million)
+                               :output-token-cost (/ 75.0 one-million)}
             "claude-3-5-haiku-latest" {:tools true
-                                       :web-search true}} ;; + ollama local models + custom provider models
+                                       :web-search true
+                                       :input-token-cost (/ 0.8 one-million)
+                                       :output-token-cost (/ 4.0 one-million)}} ;; + ollama local models + custom provider models
    :mcp-clients {}})
 
 (defonce db* (atom initial-db))

src/eca/features/chat.clj

@@ -10,7 +10,7 @@
    [eca.llm-api :as llm-api]
    [eca.logger :as logger]
    [eca.messenger :as messenger]
-   [eca.shared :as shared]))
+   [eca.shared :as shared :refer [assoc-some]]))
 
 (set! *warn-on-reflection* true)
 
@@ -79,6 +79,15 @@
 (defn ^:private tool-name->origin [name all-tools]
   (:origin (first (filter #(= name (:name %)) all-tools))))
 
+(defn ^:private tokens->cost [input-tokens output-tokens model db]
+  (let [normalized-model (if (string/includes? model "/")
+                           (last (string/split model #"/"))
+                           model)
+        {:keys [input-token-cost output-token-cost]} (get-in db [:models normalized-model])]
+    (when (and input-token-cost output-token-cost)
+      (format "%.2f" (+ (* input-tokens input-token-cost)
+                        (* output-tokens output-token-cost))))))
+
 (defn prompt
   [{:keys [message model behavior contexts chat-id request-id]}
    db*
@@ -119,7 +128,10 @@
         received-msgs* (atom "")
         tool-call-args-by-id* (atom {})
         add-to-history! (fn [msg]
-                          (swap! db* update-in [:chats chat-id :messages] (fnil conj []) msg))]
+                          (swap! db* update-in [:chats chat-id :messages] (fnil conj []) msg))
+        sum-sesison-tokens! (fn [input-tokens output-tokens]
+                              (swap! db* update-in [:chats chat-id :total-input-tokens] (fnil + 0) input-tokens)
+                              (swap! db* update-in [:chats chat-id :total-output-tokens] (fnil + 0) output-tokens))]
     (messenger/chat-content-received
      messenger
      {:chat-id chat-id
@@ -178,6 +190,23 @@
                                                 (finish-chat-prompt! chat-id :idle messenger db*))
                                :finish (do
                                          (add-to-history! {:role "assistant" :content @received-msgs*})
+                                         (when-let [{:keys [output-tokens input-tokens]} (:usage msg)]
+                                           (when (and output-tokens input-tokens)
+                                             (sum-sesison-tokens! input-tokens output-tokens)
+                                             (let [db @db*
+                                                   total-input-tokens (get-in db [:chats chat-id :total-input-tokens] 0)
+                                                   total-output-tokens (get-in db [:chats chat-id :total-output-tokens] 0)]
+                                               (messenger/chat-content-received
+                                                messenger
+                                                {:chat-id chat-id
+                                                 :request-id request-id
+                                                 :role :system
+                                                 :content (assoc-some {:type :usage
+                                                                       :message-output-tokens output-tokens
+                                                                       :message-input-tokens input-tokens
+                                                                       :session-tokens (+ total-input-tokens total-output-tokens)}
+                                                                      :message-cost (tokens->cost input-tokens output-tokens chosen-model db)
+                                                                      :session-cost (tokens->cost total-input-tokens total-output-tokens chosen-model db))}))))
                                          (finish-chat-prompt! chat-id :idle messenger db*))))
       :on-prepare-tool-call (fn [{:keys [id name arguments-text]}]
                               (assert-chat-not-stopped! chat-id db* messenger)

src/eca/llm_providers/anthropic.clj

@@ -138,6 +138,8 @@
                                                  :on-error on-error
                                                  :on-response handle-response}))))
                               "end_turn" (on-message-received {:type :finish
+                                                               :usage {:input-tokens (-> data :usage :input_tokens)
+                                                                       :output-tokens (-> data :usage :output_tokens)}
                                                                :finish-reason (-> data :delta :stop_reason)})
                               "max_tokens" (on-message-received {:type :limit-reached
                                                                  :tokens (:usage data)})

src/eca/llm_providers/openai.clj

@@ -127,6 +127,8 @@
             "response.completed"
             (when-not (= "function_call" (-> data :response :output last :type))
               (on-message-received {:type :finish
+                                    :usage {:input-tokens (-> data :response :usage :input_tokens)
+                                            :output-tokens (-> data :response :usage :output_tokens)}
                                     :finish-reason (-> data :response :status)}))
             nil))]
     (base-completion-request!

src/eca/shared.clj

@@ -20,8 +20,21 @@
 
 (defn deep-merge [v & vs]
   (letfn [(rec-merge [v1 v2]
-                     (if (and (map? v1) (map? v2))
-                       (merge-with deep-merge v1 v2)
-                       v2))]
+            (if (and (map? v1) (map? v2))
+              (merge-with deep-merge v1 v2)
+              v2))]
     (when (some identity vs)
       (reduce #(rec-merge %1 %2) v vs))))
+
+(defn assoc-some
+  "Assoc[iate] if the value is not nil. "
+  ([m k v]
+   (if (nil? v) m (assoc m k v)))
+  ([m k v & kvs]
+   (let [ret (assoc-some m k v)]
+     (if kvs
+       (if (next kvs)
+         (recur ret (first kvs) (second kvs) (nnext kvs))
+         (throw (IllegalArgumentException.
+                 "assoc-some expects even number of arguments after map/vector, found odd number")))
+       ret))))