feat(vscode): add compaction model setting

.changeset/quiet-compaction-model.md

@@ -0,0 +1,5 @@
+---
+"kilo-code": minor
+---
+
+Choose a separate model for conversation compaction in Context settings.

packages/kilo-vscode/tests/unit/config-utils.test.ts

@@ -392,6 +392,27 @@ describe("ConfigState", () => {
     })
   })
 
+  it("sets and clears the compaction model without changing other settings", () => {
+    const s = new ConfigState()
+    const cfg: Config = {
+      model: "kilo/openai/gpt-4.1",
+      agent: { compaction: { prompt: "Keep task details" }, code: { model: "kilo/openai/gpt-4.1" } },
+    }
+    const model = "kilo/anthropic/claude-haiku-4-5"
+    s.handleConfigLoaded(cfg)
+    s.updateConfig({ agent: { compaction: { model } } })
+
+    expect(s.config).toEqual({
+      ...cfg,
+      agent: { ...cfg.agent, compaction: { prompt: "Keep task details", model } },
+    })
+    s.updateConfig({ agent: { compaction: { model: null } } })
+
+    expect(s.config).toEqual(cfg)
+    expect(s.draft.agent?.compaction?.model).toBeNull()
+    expect(configUnsetPaths(s.draft)).toEqual([["agent", "compaction", "model"]])
+  })
+
   describe("agent permission patches", () => {
     it("merges nested per-agent permission patches into existing rules", () => {
       const s = new ConfigState()

packages/kilo-vscode/webview-ui/src/components/settings/ContextTab.tsx

@@ -8,6 +8,8 @@ import { IconButton } from "@kilocode/kilo-ui/icon-button"
 import { useConfig } from "../../context/config"
 import { useLanguage } from "../../context/language"
 import { useMemory } from "../../context/memory"
+import { parseModelString } from "../../../../src/shared/provider-model"
+import { ModelSelectorBase } from "../shared/ModelSelector"
 import SettingsRow from "./SettingsRow"
 
 const ContextTab: Component = () => {
@@ -139,6 +141,24 @@ const ContextTab: Component = () => {
             {language.t("settings.context.autoCompaction.title")}
           </Switch>
         </SettingsRow>
+        <SettingsRow
+          title={language.t("settings.context.compactionModel.title")}
+          description={language.t("settings.context.compactionModel.description")}
+        >
+          <ModelSelectorBase
+            value={parseModelString(config().agent?.compaction?.model ?? undefined)}
+            onSelect={(providerID, modelID) =>
+              updateConfig({
+                agent: { compaction: { model: providerID && modelID ? `${providerID}/${modelID}` : null } },
+              })
+            }
+            placement="bottom-start"
+            allowClear
+            clearLabel={language.t("settings.context.compactionModel.useChatModel")}
+            label={language.t("settings.context.compactionModel.title")}
+            description={language.t("settings.context.compactionModel.description")}
+          />
+        </SettingsRow>
         <SettingsRow
           title={language.t("settings.context.compactionLimit.title")}
           description={language.t("settings.context.compactionLimit.description")}

packages/kilo-vscode/webview-ui/src/i18n/de.ts

@@ -1086,6 +1086,10 @@ export const dict = {
   "settings.context.autoCompaction.title": "Automatische Komprimierung",
   "settings.context.autoCompaction.description": "Kontext automatisch komprimieren, bevor er das Limit erreicht",
   "settings.context.compaction.title": "Komprimierung",
+  "settings.context.compactionModel.title": "Komprimierungsmodell",
+  "settings.context.compactionModel.description":
+    "Modell für die automatische und manuelle Komprimierung. Leer lassen, um das Chatmodell zu verwenden. Kosten, Geschwindigkeit und Zusammenfassungsqualität hängen vom Modell ab.",
+  "settings.context.compactionModel.useChatModel": "Chatmodell verwenden",
   "settings.context.compactionLimit.title": "Limit für automatische Komprimierung",
   "settings.context.compactionLimit.description":
     "Komprimieren, wenn der Kontext diesen Prozentsatz des Modellfensters erreicht. Leer lassen, um nur den Sicherheitspuffer zu verwenden.",

packages/kilo-vscode/webview-ui/src/i18n/en.ts

@@ -1038,6 +1038,10 @@ export const dict = {
   "settings.context.autoCompaction.title": "Auto Compaction",
   "settings.context.autoCompaction.description": "Automatically compact context before it reaches the limit",
   "settings.context.compaction.title": "Compaction",
+  "settings.context.compactionModel.title": "Compaction model",
+  "settings.context.compactionModel.description":
+    "Model used for automatic and manual compaction. Leave unset to use the chat model. Cost, speed, and summary quality depend on the model.",
+  "settings.context.compactionModel.useChatModel": "Use chat model",
   "settings.context.compactionLimit.title": "Auto Compaction Limit",
   "settings.context.compactionLimit.description":
     "Compact when context reaches this percentage of the model window. Leave blank to use the safety buffer only.",