gemcode 0.4.30__tar.gz → 0.4.32__tar.gz

{gemcode-0.4.30/src/gemcode.egg-info → gemcode-0.4.32}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.4.30
+Version: 0.4.32
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License

{gemcode-0.4.30 → gemcode-0.4.32}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gemcode"
-version = "0.4.30"
+version = "0.4.32"
 description = "Local-first coding agent on Google Gemini + ADK"
 readme = "README.md"
 requires-python = ">=3.11"

{gemcode-0.4.30 → gemcode-0.4.32}/src/gemcode/thinking.py

@@ -96,8 +96,16 @@ def build_thinking_config(cfg: GemCodeConfig) -> Optional[types.ThinkingConfig]:
     )
 
   # If the user only wants thought summaries, we can set include_thoughts
-  # without forcing a budget/level.
+  # without forcing a budget/level (Gemini 2.5). Gemini 3.x also needs an
+  # explicit thinking_level or the model may consume thought tokens without
+  # streaming readable summaries to the client.
   if include:
+    if not is_25:
+      show_full = bool(getattr(cfg, "show_full_thinking", False))
+      level = "high" if show_full else "medium"
+      if _is_gemini_3_pro(model_id) and level == "minimal":
+        level = "low"
+      return types.ThinkingConfig(include_thoughts=True, thinking_level=level)
     return types.ThinkingConfig(include_thoughts=True)
 
   # Otherwise: familiar auto mapping based on model_mode.

{gemcode-0.4.30 → gemcode-0.4.32}/src/gemcode/web/sse_adapter.py

@@ -53,7 +53,7 @@ def extract_parts_from_event(event: Any) -> tuple[str, str]:
       t = getattr(p, "text", None)
       if not isinstance(t, str) or not t:
         continue
-      if getattr(p, "thought", None):
+      if getattr(p, "thought", None) is True:
         thought_parts.append(t)
       else:
         final_parts.append(t)
@@ -139,9 +139,13 @@ async def _emit_text_delta(index: int, delta: str) -> None:
 
 
 async def _emit_thinking_delta(delta: str) -> None:
+  """Stream thought text in small pieces so the UI can show it before the answer."""
   if not delta:
     return
-  _sse_emit({"type": "thinking", "content": delta})
+  chunk_size = int(os.environ.get("GEMCODE_WEB_THINKING_CHUNK", "24"))
+  for piece in _iter_chunks(delta, max(1, chunk_size)):
+    _sse_emit({"type": "thinking", "content": piece})
+    await asyncio.sleep(0.008)
 
 
 def _emit_status(phase: str, *, message: str = "", elapsed_s: float | None = None) -> None:
@@ -1035,9 +1039,16 @@ def _apply_web_request_options(cfg: GemCodeConfig, req: dict[str, Any], workspac
     elif tm == "brief":
       cfg.disable_thinking = False
       cfg.show_full_thinking = False
+      # Web UI "brief" should still request thought summaries so the UI can
+      # render a streaming/collapsible "Thinking" block.
+      cfg.include_thought_summaries = True
     elif tm == "auto":
       cfg.disable_thinking = False
       cfg.show_full_thinking = False
+      # Treat web "auto" as "show brief thinking when available".
+      # Without this, Gemini can enter a thinking phase but the adapter
+      # won't receive any thought content to stream via SSE.
+      cfg.include_thought_summaries = True
 
   caps = req.get("capabilities")
   if isinstance(caps, dict):

{gemcode-0.4.30 → gemcode-0.4.32/src/gemcode.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.4.30
+Version: 0.4.32
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License

{gemcode-0.4.30 → gemcode-0.4.32}/tests/test_thinking_config.py

@@ -37,6 +37,26 @@ def test_include_thought_summaries_only_gemini25(tmp_path) -> None:
   assert thinking_cfg == types.ThinkingConfig(include_thoughts=True)
 
 
+def test_include_thought_summaries_gemini3_sets_level(tmp_path) -> None:
+  cfg = GemCodeConfig(project_root=tmp_path)
+  cfg.model = "gemini-3.1-pro-preview"
+  cfg.disable_thinking = False
+  cfg.include_thought_summaries = True
+  cfg.show_full_thinking = False
+  thinking_cfg = build_thinking_config(cfg)
+  assert thinking_cfg == types.ThinkingConfig(
+    include_thoughts=True,
+    thinking_level="medium",
+  )
+
+  cfg.show_full_thinking = True
+  thinking_cfg = build_thinking_config(cfg)
+  assert thinking_cfg == types.ThinkingConfig(
+    include_thoughts=True,
+    thinking_level="high",
+  )
+
+
 def test_thinking_level_is_ignored_for_gemini25(tmp_path) -> None:
   cfg = GemCodeConfig(project_root=tmp_path)
   cfg.model = "gemini-2.5-flash"