From 682e9158572b61fe40f06dc3e699187a0928ab37 Mon Sep 17 00:00:00 2001
From: dst1213 <dst1213@126.com>
Date: Fri, 8 May 2026 20:46:48 +0800
Subject: [PATCH] fix: retry quota errors returned as HTTP 400

---
 crates/tui/src/client.rs         |  4 ----
 crates/tui/src/llm_client/mod.rs | 19 ++++++++++++++++++-
 2 files changed, 18 insertions(+), 5 deletions(-)

diff --git a/crates/tui/src/client.rs b/crates/tui/src/client.rs
index 5f2c80fa..376ac940 100644
--- a/crates/tui/src/client.rs
+++ b/crates/tui/src/client.rs
@@ -654,10 +654,6 @@ impl DeepSeekClient {
                     if status.is_success() {
                         return Ok(response);
                     }
-                    let retryable = status.as_u16() == 429 || status.is_server_error();
-                    if !retryable {
-                        return Ok(response);
-                    }
                     let retry_after = extract_retry_after(response.headers());
                     let body = bounded_error_text(response, ERROR_BODY_MAX_BYTES).await;
                     Err(LlmError::from_http_response_with_retry_after(
diff --git a/crates/tui/src/llm_client/mod.rs b/crates/tui/src/llm_client/mod.rs
index 009f701b..91cdf05c 100644
--- a/crates/tui/src/llm_client/mod.rs
+++ b/crates/tui/src/llm_client/mod.rs
@@ -202,7 +202,16 @@ impl LlmError {
             400 => {
                 // Classify 400 errors by examining the response body
                 let body_lower = body.to_lowercase();
-                if body_lower.contains("context_length")
+                if body_lower.contains("insufficientquota")
+                    || body_lower.contains("insufficient_quota")
+                    || body_lower.contains("exceeded your current quota")
+                    || body_lower.contains("quota exceeded")
+                {
+                    LlmError::RateLimited {
+                        message: body.to_string(),
+                        retry_after: None,
+                    }
+                } else if body_lower.contains("context_length")
                     || body_lower.contains("token")
                     || body_lower.contains("too long")
                     || body_lower.contains("maximum")
@@ -846,6 +855,14 @@ mod tests {
         let err = LlmError::from_http_response(400, "context_length_exceeded");
         assert!(matches!(err, LlmError::ContextLengthError(_)));
 
+        // Some OpenAI-compatible gateways return quota/rate-limit errors as HTTP 400.
+        let err = LlmError::from_http_response(
+            400,
+            r#"{"error":{"code":"insufficientquota","message":"You exceeded your current quota"}}"#,
+        );
+        assert!(matches!(err, LlmError::RateLimited { .. }));
+        assert!(err.is_retryable());
+
         // Content policy
         let err = LlmError::from_http_response(400, "content_policy_violation");
         assert!(matches!(err, LlmError::ContentPolicyError(_)));