Fixes #4529 improves retries w/ Copilot models

Author: eamodio

CHANGELOG.md

@@ -27,6 +27,10 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/) and this p
 - Changes branch favoriting to apply to both local and remote branch pairs ([#4497](https://github.com/gitkraken/vscode-gitlens/issues/4497))
 - Supports opening an explain summary document before summary content is generated ([#4328](https://github.com/gitkraken/vscode-gitlens/issues/4328))
 
+### Fixed
+
+- Fixes issues with handling token limits when using Copilot models ([#4529](https://github.com/gitkraken/vscode-gitlens/issues/4529))
+
 ## [17.3.4] - 2025-08-11
 
 ### Added

src/plus/ai/vscodeProvider.ts

@@ -129,16 +129,20 @@ export class VSCodeAIProvider implements AIProvider<typeof provider.id> {
 
 				if (ex instanceof Error && 'cause' in ex && ex.cause instanceof Error) {
 					message += `\n${ex.cause.message}`;
+				}
 
-					if (ex.cause.message.includes('exceeds token limit')) {
-						if (retries++ < 2) {
-							maxInputTokens -= 500 * retries;
-							continue;
-						}
+				if (message.includes('exceeds token limit')) {
+					if (++retries <= 3) {
+						// Reduce by 10%, then 25%, then 50% on retries 1, 2, 3 respectively
+						const reductionPercents = [0, 0.1, 0.25, 0.5] as const;
+						const reduction = reductionPercents[retries] ?? 0.5;
 
-						Logger.error(ex, scope, `Unable to ${getActionName(action)}: (${model.provider.name})`);
-						throw new AIError(AIErrorReason.RequestTooLarge, ex);
+						maxInputTokens -= Math.min(5000 * retries, maxInputTokens * reduction);
+						continue;
 					}
+
+					Logger.error(ex, scope, `Unable to ${getActionName(action)}: (${model.provider.name})`);
+					throw new AIError(AIErrorReason.RequestTooLarge, ex);
 				}
 
 				Logger.error(ex, scope, `Unable to ${getActionName(action)}: (${model.provider.name})`);