fix(ai): detect DS4 context overflow errors

closes #6262

packages/ai/CHANGELOG.md

@@ -8,6 +8,7 @@
 - Fixed OpenAI Codex user-agent construction to synchronously load Node OS metadata, avoiding a startup race that could report `pi (browser)` in Node/Bun.
 - Fixed Fireworks GLM 5.2 Fast to use the OpenAI-compatible endpoint and `thinkingLevelMap`, aligning it with GLM 5.2 ([#6195](https://github.com/earendil-works/pi/issues/6195)).
 - Fixed Amazon Bedrock prompt-cache points for Claude Fable 5 and Claude Sonnet 5 ([#6235](https://github.com/earendil-works/pi/issues/6235)).
+- Fixed DS4 server context overflow detection for `Prompt has ... tokens, but the configured context size is ... tokens` errors ([#6262](https://github.com/earendil-works/pi/issues/6262)).
 
 ### Added
 

packages/ai/src/utils/overflow.ts

@@ -24,6 +24,7 @@ import type { AssistantMessage } from "../types.ts";
  * - GitHub Copilot: "prompt token count of X exceeds the limit of Y"
  * - MiniMax: "invalid params, context window exceeds limit"
  * - Kimi For Coding: "Your request exceeded model token limit: X (requested: Y)"
+ * - DS4: "Prompt has X tokens, but the configured context size is Y tokens"
  * - Cerebras: "400/413 status code (no body)"
  * - Mistral: "Prompt contains X tokens ... too large for model with Y maximum context length"
  * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow
@@ -50,6 +51,7 @@ const OVERFLOW_PATTERNS = [
 	/context window exceeds limit/i, // MiniMax
 	/exceeded model token limit/i, // Kimi For Coding
 	/too large for model with \d+ maximum context length/i, // Mistral
+	/prompt has [\d,]+ tokens?, but the configured context size is [\d,]+ tokens?/i, // DS4 server
 	/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text
 	/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error
 	/context[_ ]length[_ ]exceeded/i, // Generic fallback
@@ -98,6 +100,7 @@ const NON_OVERFLOW_PATTERNS = [
  * - llama.cpp: "exceeds the available context size"
  * - LM Studio: "greater than the context length"
  * - Kimi For Coding: "exceeded model token limit: X (requested: Y)"
+ * - DS4: "Prompt has X tokens, but the configured context size is Y tokens"
  *
  * **Unreliable detection:**
  * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),

packages/ai/test/overflow.test.ts

@@ -63,6 +63,18 @@ describe("isContextOverflow", () => {
 		expect(isContextOverflow(message, 131072)).toBe(true);
 	});
 
+	it("detects DS4 configured context size errors", () => {
+		const message = createErrorMessage(
+			"400 Prompt has 256468 tokens, but the configured context size is 256000 tokens",
+		);
+		expect(isContextOverflow(message, 256000)).toBe(true);
+
+		const commaMessage = createErrorMessage(
+			"Prompt has 5,958,968 tokens, but the configured context size is 256,000 tokens",
+		);
+		expect(isContextOverflow(commaMessage, 256000)).toBe(true);
+	});
+
 	it("does not treat generic non-overflow Ollama errors as overflow", () => {
 		const message = createErrorMessage("500 `model runner crashed unexpectedly`");
 		expect(isContextOverflow(message, 32768)).toBe(false);