gemcode 0.3.40__tar.gz → 0.3.42__tar.gz

{gemcode-0.3.40/src/gemcode.egg-info → gemcode-0.3.42}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.3.40
+Version: 0.3.42
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License

{gemcode-0.3.40 → gemcode-0.3.42}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gemcode"
-version = "0.3.40"
+version = "0.3.42"
 description = "Local-first coding agent on Google Gemini + ADK"
 readme = "README.md"
 requires-python = ">=3.11"

{gemcode-0.3.40 → gemcode-0.3.42}/src/gemcode/agent.py

@@ -400,7 +400,7 @@ You run locally via the GemCode CLI. You are the same agent the user launched
 
 ### Intent-aware workflow
 
-Before you see this message, a lightweight LLM classifier already determined the user's intent
+Before you see this message, a **gemini-2.5-flash-lite** intent classifier already determined the user's intent
 and stored it in session state as `_gemcode_intent`. Read it and adapt your behaviour:
 
 | `_gemcode_intent` | Meaning | What to do |

{gemcode-0.3.40 → gemcode-0.3.42}/src/gemcode/intent_classifier.py

@@ -1,7 +1,7 @@
 """
 LLM-based intent pre-classifier.
 
-Before each user turn a lightweight Gemini call (gemini-2.0-flash-lite by
+Before each user turn a lightweight Gemini call (gemini-2.5-flash-lite by
 default) classifies the message into one of five intents.  The TUI / CLI
 can then:
 
@@ -10,7 +10,7 @@ can then:
     workflow automatically without re-classifying
 
 Configure:
-  GEMCODE_INTENT_MODEL  Override the classifier model (default: gemini-2.0-flash-lite)
+  GEMCODE_INTENT_MODEL  Override the classifier model (default: gemini-2.5-flash-lite)
   GEMCODE_INTENT_CLASSIFY_ENABLED  Set "0" to disable (falls back to main agent)
 """
 from __future__ import annotations
@@ -41,6 +41,20 @@ INTENT_DESCRIPTIONS: dict[str, str] = {
     INTENT_ANALYSIS:         "Systematic audit or summarisation — thorough tool sweep, then synthesise.",
 }
 
+# How the intent was determined (for TUI suffix)
+SOURCE_LOCAL = "local"   # obvious greeting / heuristic, no classifier API call
+SOURCE_LLM = "llm"       # gemini-2.5-flash-lite classifier
+SOURCE_OFF = "off"       # classifier disabled
+
+
+def show_intentifying_line(source: str) -> bool:
+    """
+    Whether to show the minimal ``∴ Intentifying…`` line in the TUI (same lane
+    as ``∴ Thinking``).  Hidden only when intent classification is disabled
+    (``SOURCE_OFF``).
+    """
+    return source != SOURCE_OFF
+
 # ── Prompts ───────────────────────────────────────────────────────────────────
 _CLASSIFY_PROMPT = """\
 Classify the user message into exactly ONE intent label from the list below.
@@ -67,7 +81,7 @@ _GREETING_SYSTEM = (
 )
 
 _CLASSIFIER_MODEL_ENV = "GEMCODE_INTENT_MODEL"
-_DEFAULT_CLASSIFIER_MODEL = "gemini-2.0-flash-lite"
+_DEFAULT_CLASSIFIER_MODEL = "gemini-2.5-flash-lite"
 
 # Single-word / very-short messages that are unambiguously greetings —
 # checked locally before spending an API call on the classifier.
@@ -98,27 +112,24 @@ def _get_api_key() -> str:
     )
 
 
-async def classify_intent(message: str) -> str:
+async def classify_intent_with_source(message: str) -> tuple[str, str]:
     """
-    Classify a user message using a lightweight Gemini call.
-
-    Returns one of: GREETING, CONCEPT, PROJECT_QUESTION, ENGINEERING_TASK, ANALYSIS.
-    Falls back to ENGINEERING_TASK on any error (the main agent handles all real
-    tasks safely with that default).
+    Classify a user message; return (intent, source).
 
-    Classification is disabled when GEMCODE_INTENT_CLASSIFY_ENABLED=0.
+    ``source`` is one of: ``SOURCE_LOCAL`` (heuristic, no API call),
+    ``SOURCE_LLM`` (classifier model), ``SOURCE_OFF`` (classifier disabled).
     """
     if not _classifier_enabled():
-        return INTENT_ENGINEERING_TASK
+        return INTENT_ENGINEERING_TASK, SOURCE_OFF
 
     stripped = (message or "").strip()
     if not stripped:
-        return INTENT_GREETING
+        return INTENT_GREETING, SOURCE_LOCAL
 
     # Fast local check for unambiguously short greetings — saves an API round-trip.
     lower = stripped.lower()
     if lower in _OBVIOUS_GREETINGS or (len(lower) <= 3 and lower.isalpha()):
-        return INTENT_GREETING
+        return INTENT_GREETING, SOURCE_LOCAL
 
     try:
         import google.genai as genai
@@ -136,14 +147,28 @@ async def classify_intent(message: str) -> str:
         label = (resp.text or "").strip().upper()
         # Exact match first
         if label in _VALID_INTENTS:
-            return label
+            return label, SOURCE_LLM
         # Partial match (model may return extra punctuation or lowercase)
         for key in _VALID_INTENTS:
             if key in label:
-                return key
-        return INTENT_ENGINEERING_TASK
+                return key, SOURCE_LLM
+        return INTENT_ENGINEERING_TASK, SOURCE_LLM
     except Exception:
-        return INTENT_ENGINEERING_TASK
+        return INTENT_ENGINEERING_TASK, SOURCE_LLM
+
+
+async def classify_intent(message: str) -> str:
+    """
+    Classify a user message using a lightweight Gemini call.
+
+    Returns one of: GREETING, CONCEPT, PROJECT_QUESTION, ENGINEERING_TASK, ANALYSIS.
+    Falls back to ENGINEERING_TASK on any error (the main agent handles all real
+    tasks safely with that default).
+
+    Classification is disabled when GEMCODE_INTENT_CLASSIFY_ENABLED=0.
+    """
+    intent, _ = await classify_intent_with_source(message)
+    return intent
 
 
 async def generate_greeting_reply(message: str) -> str:

{gemcode-0.3.40 → gemcode-0.3.42}/src/gemcode/pricing.py

@@ -16,6 +16,7 @@ from __future__ import annotations
 _PRICING_TABLE: dict[str, tuple[float, float]] = {
     # ── Gemini 2.5 ──────────────────────────────────────────────────────────
     "gemini-2.5-pro":         (3.50, 10.50),   # standard context ≤200k
+    "gemini-2.5-flash-lite":  (0.10,  0.40),   # lowest-cost 2.5 (must precede flash)
     "gemini-2.5-flash":       (0.15,  0.60),   # standard context
     "gemini-2.5-flash-8b":    (0.037, 0.15),   # 8B lite variant
     # ── Gemini 2.0 ──────────────────────────────────────────────────────────

{gemcode-0.3.40 → gemcode-0.3.42}/src/gemcode/tui/scrollback.py

@@ -516,28 +516,30 @@ async def run_gemcode_scrollback_tui(
       prompt = slash.model_prompt or prompt
 
     # ── LLM intent pre-classifier ────────────────────────────────────────────
-    # A lightweight Gemini call (gemini-2.0-flash-lite) classifies the message
-    # before the main agent runs.  Greetings are short-circuited entirely:
-    # we generate a natural reply directly and skip the main agent, so the
-    # user gets an instant response with zero unnecessary tool calls.
-    # For all other intents the classified label is stored in session state
-    # so the main agent can adapt its workflow without re-classifying.
+    # gemini-2.5-flash-lite classifies the message (same lane as Thinking)
     try:
       from gemcode.intent_classifier import (
-          classify_intent,
+          classify_intent_with_source,
           generate_greeting_reply,
+          show_intentifying_line,
           INTENT_GREETING,
           INTENT_DESCRIPTIONS,
       )
-      _intent = await classify_intent(prompt)
+      _intent, _intent_src = await classify_intent_with_source(prompt)
+      if show_intentifying_line(_intent_src):
+        print(
+            f"  \u23bf  {ansi.dim}\u2234 Intentifying\u2026{ansi.reset}"
+        )
+        print("")
 
       if _intent == INTENT_GREETING:
-        # Fast path: generate a warm reply with the lightweight model and skip
-        # the main agent entirely — no tool calls, no spinner, instant UX.
-        _reply = await generate_greeting_reply(prompt)
+        _start_anim("Replying\u2026")
+        try:
+          _reply = await generate_greeting_reply(prompt)
+        finally:
+          _stop_anim()
         print(f"  \u23bf  {ansi.bold}GemCode{ansi.reset}:")
         console.print(_RichPadding(_RichMarkdown(_reply), (0, 0, 0, 4)))
-        # Print minimal footer (no token stats since we bypassed the main model)
         print("")
         if os.environ.get("GEMCODE_TUI_TURN_RULE", "1").lower() in (
             "1", "true", "yes", "on"

{gemcode-0.3.40 → gemcode-0.3.42/src/gemcode.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.3.40
+Version: 0.3.42
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License