@ssweens/pi-vertex 1.1.3 → 1.1.4

package/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to this project will be documented in this file.
 
+## [1.1.4] - 2026-03-30
+### Fixed
+- Removed error message override for `400 (no body)` responses from Vertex MaaS models. The original message now passes through to `isContextOverflow()` which already handles this pattern, enabling proper auto-compact instead of showing a raw error to the user.
+- Use `zai` thinking format for `zai-org` publisher models (GLM-5). Previously using `openai` format which never sent `enable_thinking`, causing intermittent 400 errors from the ZAI API.
+
 ## [1.1.3] - 2026-03-26
 ### Fixed
 - Hardened Claude-on-Vertex replay for mid-session model switching (tool ID normalization, tool result adjacency, thinking signature validation).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ssweens/pi-vertex",
-  "version": "1.1.3",
+  "version": "1.1.4",
   "description": "Google Vertex AI provider for Pi coding agent - supports Gemini, Claude, and all MaaS models",
   "type": "module",
   "main": "index.ts",

package/streaming/maas.ts

@@ -364,7 +364,7 @@ export function streamMaaS(
           supportsDeveloperRole: false,
           supportsReasoningEffort: false,
           maxTokensField: "max_tokens",
-          thinkingFormat: model.publisher === "qwen" ? "qwen" : "openai",
+          thinkingFormat: model.publisher === "qwen" ? "qwen" : model.publisher === "zai-org" ? "zai" : "openai",
         },
       };
 
@@ -381,9 +381,6 @@ export function streamMaaS(
         if ("error" in event && event.error && typeof event.error === "object") {
           const err = event.error as any;
           err.model = model.id;
-          if (typeof err.errorMessage === "string" && /^400\s*(status code)?\s*\(no body\)/i.test(err.errorMessage)) {
-            err.errorMessage = `Vertex MaaS HTTP 400 (no body) for model "${apiModelId}". Not automatically treated as context overflow.`;
-          }
         }
         stream.push(event);
       }