@vietor/agent-core 0.7.4 → 0.7.5

package/README.md

@@ -46,8 +46,10 @@ const session = await createSession({
 
 session.onEvent((e) => {
   if (e.type === "assistant_delta") process.stdout.write(e.text);
-  else if (e.type === "run_metrics")
-    console.log(`tokens: ${e.inputTokens} prompt / ${e.outputTokens} completion`);
+  else if (e.type === "run_metrics") {
+    const inputTokens = e.cacheInputTokens + e.missInputTokens;
+    console.log(`tokens: ${inputTokens} input / ${e.outputTokens} output`);
+  }
 });
 
 const result = await session.prompt("What files are in the current directory?");
@@ -219,7 +221,8 @@ interface RunMetrics {
   elapsed: number;         // seconds since the current prompt started
   thinkingElapsed: number; // seconds before the first assistant text token (incl. thinking/tools)
   replyElapsed: number;    // seconds after the first assistant text token (incl. later tool rounds)
-  inputTokens: number;     // cumulative input (prompt) tokens for the current run
+  cacheInputTokens: number; // cumulative cached input (prompt) tokens for the current run
+  missInputTokens: number; // cumulative non-cached input (prompt) tokens for the current run
   outputTokens: number;    // cumulative output (completion) tokens for the current run
 }
 ```

package/dist/llm/anthropic.js

@@ -44,7 +44,7 @@ export class AnthropicAdapter extends BaseAdapter {
         const stream = this.client.messages.stream(params, { signal: opts.signal });
         if (opts.onUsage)
             stream.on("streamEvent", (e) => { if (e.type === "message_start")
-                opts.onUsage(e.message.usage.input_tokens, 0); });
+                opts.onUsage(0, e.message.usage.input_tokens, 0); });
         if (opts.onDelta)
             stream.on("text", (delta) => opts.onDelta(delta));
         if (opts.onThinking)
@@ -53,7 +53,8 @@ export class AnthropicAdapter extends BaseAdapter {
             stream.on("contentBlock", (block) => { if (block.type === "tool_use")
                 opts.onToolCall(); });
         const final = await stream.finalMessage();
-        opts.onUsage?.(final.usage.input_tokens, final.usage.output_tokens);
+        const cacheTokens = (final.usage.cache_read_input_tokens ?? 0) + (final.usage.cache_creation_input_tokens ?? 0);
+        opts.onUsage?.(cacheTokens, final.usage.input_tokens, final.usage.output_tokens);
         const thinking = [];
         let text = "";
         const toolCalls = [];

package/dist/llm/completions.js

@@ -31,7 +31,8 @@ export class CompletionsAdapter extends BaseAdapter {
         const stream = await this.client.chat.completions.create(params, { signal });
         for await (const chunk of stream) {
             if (chunk.usage) {
-                onUsage?.(chunk.usage.prompt_tokens ?? 0, chunk.usage.completion_tokens ?? 0);
+                const cached = chunk.usage.prompt_tokens_details?.cached_tokens ?? 0;
+                onUsage?.(cached, chunk.usage.prompt_tokens ?? 0, chunk.usage.completion_tokens ?? 0);
             }
             const delta = chunk.choices[0]?.delta;
             if (!delta)

package/dist/llm/responses.js

@@ -59,7 +59,8 @@ export class ResponsesAdapter extends BaseAdapter {
             throw new Error(`Responses API error: ${detail}`);
         }
         if (finalResponse.usage) {
-            onUsage?.(finalResponse.usage.input_tokens, finalResponse.usage.output_tokens);
+            const cacheTokens = finalResponse.usage.input_tokens_details.cached_tokens;
+            onUsage?.(cacheTokens, finalResponse.usage.input_tokens, finalResponse.usage.output_tokens);
         }
         const textParts = [];
         const toolCalls = [];

package/dist/llm/types.d.ts

@@ -23,7 +23,7 @@ export interface ChatOptions {
     onDelta?: (text: string) => void;
     onThinking?: (text: string) => void;
     onRetry?: (attempt: number, max: number, error: unknown) => void;
-    onUsage?: (inputTokens: number, outputTokens: number) => void;
+    onUsage?: (cacheInputTokens: number, missInputTokens: number, outputTokens: number) => void;
     onToolCall?: () => void;
     thinking?: boolean;
     signal?: AbortSignal;

package/dist/runtime/agent.d.ts

@@ -31,15 +31,18 @@ export declare class Agent {
     private todoSnapshot;
     private resolveSkill?;
     private onCompact?;
-    private inputTokens;
+    private cacheInputTokens;
+    private missInputTokens;
     private outputTokens;
     constructor(opts: AgentOptions);
     get contextTokens(): number;
     get usage(): {
-        inputTokens: number;
+        cacheInputTokens: number;
+        missInputTokens: number;
         outputTokens: number;
     };
     resetUsage(): void;
+    addUsage(cacheInputTokens: number, missInputTokens: number, outputTokens: number): void;
     get model(): string;
     get thinkingEffort(): import("../llm/types.js").LLMThinkingEffort;
     clear(): void;

package/dist/runtime/agent.js

@@ -18,7 +18,8 @@ export class Agent {
     todoSnapshot = [];
     resolveSkill;
     onCompact;
-    inputTokens = 0;
+    cacheInputTokens = 0;
+    missInputTokens = 0;
     outputTokens = 0;
     constructor(opts) {
         this.llm = opts.llm;
@@ -37,12 +38,18 @@ export class Agent {
         return this.conversation.getEstimatedTokens();
     }
     get usage() {
-        return { inputTokens: this.inputTokens, outputTokens: this.outputTokens };
+        return { cacheInputTokens: this.cacheInputTokens, missInputTokens: this.missInputTokens, outputTokens: this.outputTokens };
     }
     resetUsage() {
-        this.inputTokens = 0;
+        this.cacheInputTokens = 0;
+        this.missInputTokens = 0;
         this.outputTokens = 0;
     }
+    addUsage(cacheInputTokens, missInputTokens, outputTokens) {
+        this.cacheInputTokens += cacheInputTokens;
+        this.missInputTokens += missInputTokens;
+        this.outputTokens += outputTokens;
+    }
     get model() {
         return this.llm.model;
     }
@@ -207,6 +214,7 @@ export class Agent {
     }
     async chatOnce(opts, onAbort) {
         try {
+            let usage;
             const message = await withAbort(this.llm.chat({
                 messages: opts.messages,
                 tools: opts.tools,
@@ -214,12 +222,13 @@ export class Agent {
                 onDelta: (text) => opts.onEvent?.({ type: "assistant_delta", text }),
                 onThinking: (text) => opts.onEvent?.({ type: "thinking_delta", text }),
                 onRetry: (attempt, max, error) => opts.onEvent?.({ type: "retry", attempt, max, reason: toErrorMessage(error) }),
-                onUsage: (inputTokens, outputTokens) => {
-                    this.inputTokens = inputTokens;
-                    this.outputTokens = outputTokens;
+                onUsage: (cacheInputTokens, missInputTokens, outputTokens) => {
+                    usage = { cacheInputTokens, missInputTokens, outputTokens };
                 },
                 signal: opts.signal,
             }), opts.signal);
+            if (usage)
+                this.addUsage(usage.cacheInputTokens, usage.missInputTokens, usage.outputTokens);
             return { ok: true, message };
         }
         catch (e) {

package/dist/runtime/events.d.ts

@@ -3,7 +3,8 @@ export interface RunMetrics {
     elapsed: number;
     thinkingElapsed: number;
     replyElapsed: number;
-    inputTokens: number;
+    cacheInputTokens: number;
+    missInputTokens: number;
     outputTokens: number;
 }
 export declare const INITIAL_RUN_METRICS: RunMetrics;

package/dist/runtime/events.js

@@ -1 +1 @@
-export const INITIAL_RUN_METRICS = { running: false, elapsed: 0, thinkingElapsed: 0, replyElapsed: 0, inputTokens: 0, outputTokens: 0 };
+export const INITIAL_RUN_METRICS = { running: false, elapsed: 0, thinkingElapsed: 0, replyElapsed: 0, cacheInputTokens: 0, missInputTokens: 0, outputTokens: 0 };

package/dist/runtime/session.js

@@ -211,6 +211,7 @@ export class Session {
                     maxTurns: deps.maxTurns ?? DEFAULT_MAX_TURNS,
                     stallThreshold: deps.stallThreshold ?? DEFAULT_STALL_THRESHOLD,
                     contextLimit: deps.contextLimit,
+                    onUsage: (cacheInputTokens, missInputTokens, outputTokens) => this.agent.addUsage(cacheInputTokens, missInputTokens, outputTokens),
                 })(systemPrompt, task, signal),
             },
         });

package/dist/runtime/sub-agent-runner.d.ts

@@ -9,6 +9,7 @@ export interface SubAgentRunOptions {
     maxTurns: number;
     stallThreshold: number;
     contextLimit: number;
+    onUsage?: (cacheInputTokens: number, missInputTokens: number, outputTokens: number) => void;
 }
 export interface SubAgentRunResult {
     status: RunStatus;

package/dist/runtime/sub-agent-runner.js

@@ -19,6 +19,7 @@ export function createSubAgentRunner(opts) {
             contextLimit: opts.contextLimit,
         });
         const status = await subAgent.run(task, undefined, signal);
+        opts.onUsage?.(subAgent.usage.cacheInputTokens, subAgent.usage.missInputTokens, subAgent.usage.outputTokens);
         const reply = conversation.lastAssistantText() || `(sub-agent produced no final text; status ${status})`;
         const messages = status !== "ok" ? conversation.export() : [];
         return { status, reply, messages };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vietor/agent-core",
-  "version": "0.7.4",
+  "version": "0.7.5",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",