chore(showcase/google-adk): bump Gemini default to 3.1 Flash-Lite

Bumps the default Gemini model across the Google ADK showcase from
gemini-2.5-flash to gemini-3.1-flash-lite — the closest GA model in
the Gemini 3.x line (2.5 family is slated for June 2026 deprecation).

Touches agent factories (DEFAULT_MODEL, _DEFAULT_A2UI_MODEL, _SUB_MODEL),
the A2UI default-model test fixture, QA docs, the manifest description
for the reasoning demo, and prose references in shared_chat docstring +
entrypoint comment.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: tylerslaton

showcase/integrations/google-adk/entrypoint.sh

@@ -13,7 +13,7 @@ trap cleanup EXIT
 export PYTHONUNBUFFERED=1
 
 # Disable Google ADK's progressive SSE streaming feature. With it enabled,
-# Gemini 2.5-flash occasionally returns a stream whose final event is flagged
+# Gemini 3.1 Flash-Lite occasionally returns a stream whose final event is flagged
 # `partial`, which the ADK flow aborts with a "The last event is partial"
 # warning — the backend then emits no TOOL_CALL_* or TEXT_MESSAGE_* events,
 # so the tool-rendering UI is stranded and L4 smoke tests intermittently fail.

showcase/integrations/google-adk/manifest.yaml

@@ -322,7 +322,7 @@ demos:
       - src/app/api/copilotkit/route.ts
   - id: reasoning-custom
     name: "Reasoning: Custom"
-    description: Visible reasoning/thinking chain alongside the final answer (Gemini 2.5 thinking mode)
+    description: Visible reasoning/thinking chain alongside the final answer (Gemini 3.1 thinking mode)
     tags:
       - chat-ui
     route: /demos/reasoning-custom

showcase/integrations/google-adk/qa/headless-complete.md

@@ -3,7 +3,7 @@
 ## Prerequisites
 
 - Demo is deployed and accessible at `/demos/headless-complete` on the dashboard host
-- Agent backend is healthy (`/api/health`); `GOOGLE_API_KEY` is set on Railway; the ADK backend mounts the shared `_simple_chat` LlmAgent (Gemini 2.5 Flash) at `/headless_complete` — the demo relies on aimock fixtures to supply tool calls (`get_weather`, `get_stock_price`, `get_revenue_chart`) since the ADK simple agent has no real backend tools
+- Agent backend is healthy (`/api/health`); `GOOGLE_API_KEY` is set on Railway; the ADK backend mounts the shared `_simple_chat` LlmAgent (Gemini 3.1 Flash-Lite) at `/headless_complete` — the demo relies on aimock fixtures to supply tool calls (`get_weather`, `get_stock_price`, `get_revenue_chart`) since the ADK simple agent has no real backend tools
 - The demo wires `agent="headless-complete"` at `/api/copilotkit-mcp-apps` (shared with the mcp-apps cell) so the Excalidraw MCP server at `MCP_SERVER_URL || https://mcp.excalidraw.com` is available
 - Note: the demo defines headless-specific `data-testid`s in `chat/message-list.tsx`, `chat/composer.tsx`, `tools/weather-card.tsx`, `tools/stock-card.tsx`, `tools/highlight-note.tsx`, and `tools/chart-card.tsx`. Other checks rely on verbatim text, role selectors, and Tailwind utility classes
 

showcase/integrations/google-adk/qa/headless-simple.md

@@ -3,7 +3,7 @@
 ## Prerequisites
 
 - Demo is deployed and accessible at `/demos/headless-simple` on the dashboard host
-- Agent backend is healthy (`/api/health`); `GOOGLE_API_KEY` is set on Railway; the ADK backend mounts the shared `_simple_chat` LlmAgent (Gemini 2.5 Flash) at `/headless_simple`
+- Agent backend is healthy (`/api/health`); `GOOGLE_API_KEY` is set on Railway; the ADK backend mounts the shared `_simple_chat` LlmAgent (Gemini 3.1 Flash-Lite) at `/headless_simple`
 - The demo wires `agent="headless-simple"` at `/api/copilotkit` (neutral assistant cell)
 - Note: the demo source contains no `data-testid` attributes. Checks below rely on verbatim visible text, role/button selectors, and Tailwind utility-class structure
 

showcase/integrations/google-adk/qa/multimodal.md

@@ -5,7 +5,7 @@
 - Demo deployed and accessible at `/demos/multimodal`
 - Railway service `showcase-google-adk` is healthy
 - `GOOGLE_API_KEY` set on the Railway service (Gemini calls require it)
-- Agent is using a natively-multimodal Gemini model (`gemini-2.5-flash` or
+- Agent is using a natively-multimodal Gemini model (`gemini-3.1-flash-lite` or
   equivalent) — verifiable by inspecting the Railway logs for
   `src/agents/multimodal_agent.py` or by running one image round-trip
 - Sample files are bundled under `public/demo-files/`:

showcase/integrations/google-adk/src/agents/beautiful_chat_agent.py

@@ -56,7 +56,7 @@
 # Model used for the secondary A2UI planner call. Mirrors main.py so the
 # A2UI surface generated for Beautiful Chat behaves identically to the
 # Sales Pipeline demo.
-_DEFAULT_A2UI_MODEL = "gemini-2.5-flash"
+_DEFAULT_A2UI_MODEL = "gemini-3.1-flash-lite"
 
 
 def _a2ui_model() -> str:

showcase/integrations/google-adk/src/agents/main.py

@@ -36,9 +36,9 @@
 logger = logging.getLogger(__name__)
 
 # Model used for the secondary A2UI planner call. Overridable via A2UI_MODEL
-# env var. gemini-2.5-flash is cheap, fast, and supports forced tool-calling
+# env var. gemini-3.1-flash-lite is fast and supports forced tool-calling
 # via ToolConfig.function_calling_config.mode="ANY".
-_DEFAULT_A2UI_MODEL = "gemini-2.5-flash"
+_DEFAULT_A2UI_MODEL = "gemini-3.1-flash-lite"
 
 
 def _a2ui_model() -> str:

showcase/integrations/google-adk/src/agents/multimodal_agent.py

@@ -1,6 +1,6 @@
 """Agent backing the Multimodal Attachments demo.
 
-Gemini 2.5 Flash is natively multimodal — image and PDF parts forwarded
+Gemini 3.1 Flash-Lite is natively multimodal — image and PDF parts forwarded
 through ADK as `types.Part` blobs are processed by the model directly.
 No backend tools needed; the model just describes / answers questions
 about uploaded attachments.

showcase/integrations/google-adk/src/agents/registry.py

@@ -102,7 +102,7 @@ class AgentSpec:
     name="SimpleChatAgent", instruction=_SIMPLE_CHAT_INSTRUCTION
 )
 
-# Reasoning-mode agent (Gemini 2.5 thinking). One shared instance for both
+# Reasoning-mode agent (Gemini 3.1 thinking). One shared instance for both
 # reasoning demos — they differ only in frontend wiring.
 _thinking_chat = build_thinking_chat_agent(
     name="ThinkingChatAgent",

showcase/integrations/google-adk/src/agents/shared_chat.py

@@ -5,18 +5,18 @@
 (prebuilt-sidebar, prebuilt-popup, chat-slots, chat-customization-css,
 headless-simple, headless-complete, voice, frontend-tools, agentic-chat).
 
-`build_thinking_chat_agent` uses Gemini 2.5 Flash with the thinking_config
+`build_thinking_chat_agent` uses Gemini 3.1 Flash-Lite with the thinking_config
 exposed so reasoning is streamed back as `thought` parts; the v2 React core
 renders these via CopilotChatReasoningMessage.
 
 `get_model` returns a `Gemini` instance configured with the aimock proxy
 endpoint when `GOOGLE_GEMINI_BASE_URL` is set, or the default model string
 otherwise. All agent modules should call `get_model()` instead of
-hard-coding `"gemini-2.5-flash"` so Railway deployments route through
+hard-coding `"gemini-3.1-flash-lite"` so Railway deployments route through
 aimock.
 
 `stop_on_terminal_text` is the canonical after_model_callback shared by every
-registered LlmAgent. Gemini 2.5-flash does not naturally end its agentic
+registered LlmAgent. Gemini 3.1 Flash-Lite does not naturally end its agentic
 loop after a successful tool call — it keeps re-issuing the same tool. The
 callback inspects each non-partial model response and, when it contains
 text with no pending function_call, sets `_invocation_context.end_invocation
@@ -39,7 +39,7 @@
 
 logger = logging.getLogger(__name__)
 
-DEFAULT_MODEL = "gemini-2.5-flash"
+DEFAULT_MODEL = "gemini-3.1-flash-lite"
 
 
 def stop_on_terminal_text(
@@ -56,7 +56,7 @@ def stop_on_terminal_text(
        in `entrypoint.sh`).
     2. Only terminate when the final non-partial response contains TEXT
        and NO pending function_call — mixed text+function_call responses
-       (a known Gemini 2.5-flash quirk) must NOT terminate.
+       (a known Gemini Flash quirk) must NOT terminate.
     3. `_invocation_context` is an ADK private attribute; if it disappears
        in a future ADK release, log-and-degrade rather than crash the
        callback (which would stall the request).