From 69ba0da3051865bd6f60346afff26f6870e8e87b Mon Sep 17 00:00:00 2001
From: IfkumRfnl <neto_frye@mail.ru>
Date: Mon, 3 Aug 2026 05:49:31 +0000
Subject: [PATCH] fix(stream): correct cached input accounting

---
 CHANGELOG.md         |  4 ++++
 src/core.ts          | 10 +++++++---
 tests/test-stream.ts | 11 +++++++----
 3 files changed, 18 insertions(+), 7 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index df21563..e6d6c4a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,9 @@
 # Changelog
 
+## 0.4.4 - 2026-08-03
+
+- Fix cached input tokens being counted twice.
+
 ## 0.4.3 - 2026-08-02
 
 - Allow pi to start when model discovery is unavailable. The provider now caches the last successfully fetched model catalog so previously discovered Command Code models remain selectable offline; a first offline start without a cache keeps Command Code unavailable until `/reload` succeeds.
diff --git a/src/core.ts b/src/core.ts
index 301ff64..e696c86 100644
--- a/src/core.ts
+++ b/src/core.ts
@@ -388,10 +388,14 @@ export function createStreamCommandCode(deps: CoreDependencies) {
             const usage = commandCodeUsage(event)
             if (usage) {
               const details = commandCodeInputTokenDetails(usage)
-              output.usage.input = numberValue(usage.inputTokens) ?? 0
+              const totalInput = numberValue(usage.inputTokens) ?? 0
+              const input = numberValue(details?.noCacheTokens)
+              const cacheRead = numberValue(details?.cacheReadTokens) ?? 0
+              const cacheWrite = numberValue(details?.cacheWriteTokens) ?? 0
+              output.usage.input = input ?? Math.max(0, totalInput - cacheRead - cacheWrite)
               output.usage.output = numberValue(usage.outputTokens) ?? 0
-              output.usage.cacheRead = numberValue(details?.cacheReadTokens) ?? 0
-              output.usage.cacheWrite = numberValue(details?.cacheWriteTokens) ?? 0
+              output.usage.cacheRead = cacheRead
+              output.usage.cacheWrite = cacheWrite
               output.usage.totalTokens =
                 output.usage.input +
                 output.usage.output +
diff --git a/tests/test-stream.ts b/tests/test-stream.ts
index 225b52d..31e4d03 100644
--- a/tests/test-stream.ts
+++ b/tests/test-stream.ts
@@ -102,9 +102,9 @@ describe("streamCommandCode — successful streams", () => {
           type: "finish",
           finishReason: "stop",
           totalUsage: {
-            inputTokens: 5,
-            outputTokens: 2,
-            inputTokenDetails: { cacheReadTokens: 3, cacheWriteTokens: 1 },
+            inputTokens: 3124,
+            outputTokens: 15,
+            inputTokenDetails: { noCacheTokens: 52, cacheReadTokens: 3072 },
           },
         }),
       ],
@@ -134,7 +134,10 @@ describe("streamCommandCode — successful streams", () => {
       done.message.content[0]?.type === "text" ? done.message.content[0].text : "",
       "Hello",
     )
-    assert.equal(done.message.usage.totalTokens, 11)
+    assert.equal(done.message.usage.input, 52)
+    assert.equal(done.message.usage.cacheRead, 3072)
+    assert.equal(done.message.usage.cacheWrite, 0)
+    assert.equal(done.message.usage.totalTokens, 3139)
     assert.equal(calculatedUsages.length, 1)
   })