From e8d3d4340fa67d766dd435632be18aa443530b2f Mon Sep 17 00:00:00 2001
From: ycd <ycd>
Date: Tue, 8 Sep 2026 21:14:14 +0800
Subject: [PATCH] fix(providers): detect llama.cpp context size overflow in
 error classifier

---
 internal/providers/error_classify.go      | 3 +++
 internal/providers/error_classify_test.go | 8 ++++++++
 2 files changed, 11 insertions(+)

diff --git a/internal/providers/error_classify.go b/internal/providers/error_classify.go
index 61df90f3..ead5116f 100644
--- a/internal/providers/error_classify.go
+++ b/internal/providers/error_classify.go
@@ -170,6 +170,9 @@ func isContextOverflow(lower string) bool {
 		"request_too_large",         // Generic
 		"input is too long",         // DashScope
 		"exceed_context_size",       // Ollama native /api/chat 400
+		// llama.cpp / OpenAI-compat local servers return:
+		//   "request (N tokens) exceeds the available context size (M tokens)"
+		"context size",              // llama.cpp native /v1/chat/completions 400
 		"请求输入过长",                    // Chinese generic
 	)
 }
diff --git a/internal/providers/error_classify_test.go b/internal/providers/error_classify_test.go
index 6a941186..35f53840 100644
--- a/internal/providers/error_classify_test.go
+++ b/internal/providers/error_classify_test.go
@@ -151,6 +151,14 @@ func TestClassifyTooManyTokens(t *testing.T) {
 	}
 }
 
+func TestClassifyContextOverflowLlamaCpp(t *testing.T) {
+	classifier := NewDefaultClassifier()
+	result := classifier.Classify(nil, 400, `{"error":{"code":400,"message":"request (218678 tokens) exceeds the available context size (204800 tokens), try increasing it","type":"exceed_context_size_error","n_prompt_tokens":218678,"n_ctx":204800}}`)
+	if result.Kind != "context_overflow" {
+		t.Errorf("expected context_overflow for llama.cpp message, got %s", result.Kind)
+	}
+}
+
 func TestClassifyNetworkTimeoutError(t *testing.T) {
 	classifier := NewDefaultClassifier()
 	timeoutErr := &net.DNSError{