feat: bump default Gemini model to 3.5-flash

.env.example

@@ -10,7 +10,7 @@
 # LLM_PROVIDER=gemini
 
 # Model name (works with both providers). For litellm, use "provider/model" format
-# (e.g., "openai/gpt-4o", "vertex_ai/gemini-2.5-flash"). For gemini, any provider/
+# (e.g., "openai/gpt-4o", "vertex_ai/gemini-3.5-flash"). For gemini, any provider/
 # prefix is stripped automatically. Falls back to GEMINI_MODEL when not set.
 # LLM_MODEL=
 
@@ -39,8 +39,8 @@ GOOGLE_CLOUD_LOCATION=global
 # Path to a service account JSON key file. Uses Application Default Credentials (ADC).
 # GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
 
-# Model to use (default: gemini-2.5-flash)
-GEMINI_MODEL=gemini-2.5-flash
+# Model to use (default: gemini-3.5-flash)
+GEMINI_MODEL=gemini-3.5-flash
 
 # Gemini HTTP retries (google-genai SDK: exponential backoff + jitter for 429/408/5xx)
 # See: https://cloud.google.com/vertex-ai/generative-ai/docs/retry-strategy

CLAUDE.md

@@ -4,7 +4,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 ## Project Overview
 
-Red Hat Lightspeed Agent for Google Cloud — an A2A-ready (Agent-to-Agent protocol) AI agent providing access to Red Hat Insights, built on Google Agent Development Kit (ADK), Gemini 2.5 Flash, and a Red Hat Lightspeed MCP server sidecar. Integrates with Google Cloud Marketplace for provisioning, billing, and metering.
+Red Hat Lightspeed Agent for Google Cloud — an A2A-ready (Agent-to-Agent protocol) AI agent providing access to Red Hat Insights, built on Google Agent Development Kit (ADK), Gemini 3.5 Flash, and a Red Hat Lightspeed MCP server sidecar. Integrates with Google Cloud Marketplace for provisioning, billing, and metering.
 
 ## Common Commands
 
@@ -140,5 +140,5 @@ CORS → body size limits → security headers → rate limiting (60 req/min, 10
 
 All configuration is via environment variables, managed through Pydantic settings in `config/settings.py`. See `.env.example` for the complete list (30+ vars) and `docs/configuration.md` for detailed documentation.
 
-Key variables: `GOOGLE_API_KEY` (or Vertex AI), `GEMINI_MODEL` (default: `gemini-2.5-flash`), `DATABASE_URL`, `SESSION_DATABASE_URL`, `MCP_TRANSPORT_MODE`, `MCP_SERVER_URL`, `RED_HAT_SSO_CLIENT_ID`/`SECRET`. Dev-only bypasses: `SKIP_JWT_VALIDATION`, `SKIP_DCR_JWT_VALIDATION`, `SKIP_ORDER_VALIDATION`.
+Key variables: `GOOGLE_API_KEY` (or Vertex AI), `GEMINI_MODEL` (default: `gemini-3.5-flash`), `DATABASE_URL`, `SESSION_DATABASE_URL`, `MCP_TRANSPORT_MODE`, `MCP_SERVER_URL`, `RED_HAT_SSO_CLIENT_ID`/`SECRET`. Dev-only bypasses: `SKIP_JWT_VALIDATION`, `SKIP_DCR_JWT_VALIDATION`, `SKIP_ORDER_VALIDATION`.
 

EXTERNAL_SERVICES.md

@@ -13,7 +13,7 @@ The agent uses Google Gemini models by default, but Vertex AI supports deploying
 
 | Setting | Default | Description |
 |---------|---------|-------------|
-| `GEMINI_MODEL` | `gemini-2.5-flash` | Model used for agent responses |
+| `GEMINI_MODEL` | `gemini-3.5-flash` | Model used for agent responses |
 | `GOOGLE_GENAI_USE_VERTEXAI` | `false` | Switch between AI Studio and Vertex AI |
 | `GOOGLE_CLOUD_LOCATION` | `us-central1` | Region for Vertex AI |
 

deploy/cloudrun/README.md

@@ -1258,8 +1258,9 @@ gcloud run services update lightspeed-agent \
 
 **Tuning tips:**
 
-- **51,200 characters** (default, 50 KiB) is a conservative limit that keeps input tokens well within
-  Vertex AI TPM quotas for `gemini-2.5-flash`
+- **51,200 characters** (default, 50 KiB) is a conservative starting point meant to keep input
+  tokens well within your Vertex AI TPM quota — verify against the quota for whichever model
+  you configure via `GEMINI_MODEL`, since quotas and context windows vary by model
 - If you have higher TPM quotas, increase the limit to allow richer responses
 - The optimal limit depends on the model's context window and expected session length — longer
   multi-turn sessions accumulate more context, leaving less room for individual tool results.

deploy/cloudrun/service.yaml

@@ -64,7 +64,7 @@ spec:
             - name: GOOGLE_CLOUD_LOCATION
               value: "${VERTEXAI_LOCATION}"
             - name: GEMINI_MODEL
-              value: "gemini-2.5-flash"
+              value: "gemini-3.5-flash"
             # Agent Configuration
             - name: AGENT_HOST
               value: "0.0.0.0"

deploy/openshift/README.md

@@ -587,7 +587,7 @@ The MCP server runs as a sidecar container in the agent pod.
 
 | Value | Description | Default |
 |---|---|---|
-| `google.geminiModel` | Default Gemini model (used when `llm.model` is not set) | `gemini-2.5-flash` |
+| `google.geminiModel` | Default Gemini model (used when `llm.model` is not set) | `gemini-3.5-flash` |
 | `google.useVertexAI` | Use Vertex AI instead of AI Studio | `false` |
 | `google.cloudLocation` | Vertex AI region (use `global` for pay-as-you-go) | `global` |
 | `secrets.gcpServiceAccountKey` | GCP service account key JSON for Vertex AI authentication via ADC (alternative to `secrets.googleApiKey`) | `""` |
@@ -624,7 +624,7 @@ To use Vertex AI with service account authentication, set in `my-values.yaml`:
 
 ```yaml
 llm:
-  model: "vertex_ai/gemini-2.5-flash"
+  model: "vertex_ai/gemini-3.5-flash"
 
 google:
   useVertexAI: true

deploy/openshift/values.yaml

@@ -97,7 +97,7 @@ google:
   # Set to true to use Vertex AI instead of Google AI Studio
   useVertexAI: false
   cloudLocation: global
-  geminiModel: gemini-2.5-flash
+  geminiModel: gemini-3.5-flash
   httpRetry:
     attempts: 5
     initialDelay: 1.0
@@ -113,7 +113,7 @@ llm:
   # "litellm" — use any provider supported by LiteLLM (OpenAI, Anthropic, etc.).
   provider: "gemini"
   # Model name (works with both providers). For litellm, use "provider/model"
-  # format (e.g., "vertex_ai/gemini-2.5-flash"). For gemini, any provider/
+  # format (e.g., "vertex_ai/gemini-3.5-flash"). For gemini, any provider/
   # prefix is stripped automatically. Falls back to google.geminiModel when empty.
   model: ""
   # Custom API endpoint URL (litellm only, for self-hosted models).

deploy/podman/lightspeed-agent-configmap.yaml

@@ -11,7 +11,7 @@ data:
   # Google AI Configuration
   GOOGLE_GENAI_USE_VERTEXAI: "FALSE"
   GOOGLE_CLOUD_LOCATION: "global"
-  GEMINI_MODEL: "gemini-2.5-flash"
+  GEMINI_MODEL: "gemini-3.5-flash"
 
   # Red Hat SSO Configuration
   RED_HAT_SSO_ISSUER: "https://sso.redhat.com/auth/realms/redhat-external"

docs/api.md

@@ -241,7 +241,7 @@ with the [A2A protocol](https://google.github.io/A2A/) (Agent-to-Agent) for inte
 │  │                      (agents/llm_agent.py)                           │   │
 │  │                                                                      │   │
 │  │  + name: str                                                         │   │
-│  │  + model: str (e.g., "gemini-2.5-flash")                             │   │
+│  │  + model: str (e.g., "gemini-3.5-flash")                             │   │
 │  │  + instruction: str                                                  │   │
 │  │  + tools: list[BaseTool | McpToolset]                                │   │
 │  └───────────────────────────────────┬──────────────────────────────────┘   │

docs/configuration.md

@@ -13,7 +13,7 @@ The agent supports multiple LLM backends via Google ADK. By default it uses Gemi
 | Variable | Default | Description |
 |----------|---------|-------------|
 | `LLM_PROVIDER` | `gemini` | LLM provider backend: `gemini` (Google AI Studio or Vertex AI) or `litellm` (100+ providers via LiteLLM) |
-| `LLM_MODEL` | - | Model name (works with both providers). For `litellm`: uses `provider/model` format (e.g., `openai/gpt-4o`, `vertex_ai/gemini-2.5-flash`). For `gemini`: any `provider/` prefix is stripped automatically. Falls back to `GEMINI_MODEL` when not set |
+| `LLM_MODEL` | - | Model name (works with both providers). For `litellm`: uses `provider/model` format (e.g., `openai/gpt-4o`, `vertex_ai/gemini-3.5-flash`). For `gemini`: any `provider/` prefix is stripped automatically. Falls back to `GEMINI_MODEL` when not set |
 | `LLM_API_KEY` | - | API key for non-Google LLM providers (`litellm` only). Some providers also accept their own env vars (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.) |
 | `LLM_API_BASE` | - | Custom API endpoint URL (`litellm` only). For self-hosted models or proxy endpoints |
 
@@ -22,13 +22,13 @@ The agent supports multiple LLM backends via Google ADK. By default it uses Gemi
 ```bash
 # These are the defaults; no LLM_* variables required
 LLM_PROVIDER=gemini
-GEMINI_MODEL=gemini-2.5-flash
+GEMINI_MODEL=gemini-3.5-flash
 ```
 
 **Different Gemini model:**
 
 ```bash
-LLM_MODEL=gemini-2.0-flash
+LLM_MODEL=gemini-2.5-flash
 ```
 
 **OpenAI:**
@@ -80,7 +80,7 @@ These settings apply when `LLM_PROVIDER=gemini` (the default).
 | `GOOGLE_API_KEY` | - | Google AI Studio API key (required if not using Vertex AI) |
 | `GOOGLE_CLOUD_PROJECT` | - | GCP project ID (required for Vertex AI) |
 | `GOOGLE_CLOUD_LOCATION` | `global` | Vertex AI model location (use `global` for pay-as-you-go) |
-| `GEMINI_MODEL` | `gemini-2.5-flash` | Default Gemini model (used when `LLM_MODEL` is not set) |
+| `GEMINI_MODEL` | `gemini-3.5-flash` | Default Gemini model (used when `LLM_MODEL` is not set) |
 | `GOOGLE_APPLICATION_CREDENTIALS` | - | Path to GCP service account key JSON file for Vertex AI authentication (ADC) |
 | `GEMINI_HTTP_RETRY_ATTEMPTS` | `5` | Max HTTP attempts per model call (including the first). Use `1` to disable SDK retries. Aligns with [google-genai defaults](https://cloud.google.com/vertex-ai/generative-ai/docs/retry-strategy). |
 | `GEMINI_HTTP_RETRY_INITIAL_DELAY` | `1.0` | Initial backoff delay in seconds (exponential backoff with jitter). |

docs/troubleshooting.md

@@ -180,7 +180,7 @@ curl -s -o /dev/null -w "%{http_code}" \
 
 ```bash
 # Test Gemini API directly
-curl -X POST "https://generativelanguage.googleapis.com/v1/models/gemini-2.5-flash:generateContent?key=$GOOGLE_API_KEY" \
+curl -X POST "https://generativelanguage.googleapis.com/v1/models/gemini-3.5-flash:generateContent?key=$GOOGLE_API_KEY" \
   -H "Content-Type: application/json" \
   -d '{"contents":[{"parts":[{"text":"Hello"}]}]}'
 ```

src/lightspeed_agent/config/settings.py

@@ -72,7 +72,7 @@ class Settings(BaseSettings):
         description="Google Cloud location for Vertex AI (use 'global' for pay-as-you-go)",
     )
     gemini_model: str = Field(
-        default="gemini-2.5-flash",
+        default="gemini-3.5-flash",
         description="Gemini model to use",
     )
     google_application_credentials: str | None = Field(

src/lightspeed_agent/core/agent.py

@@ -251,7 +251,7 @@ def _create_model(settings: Settings) -> BaseLlm:
         return LiteLlm(**kwargs)
 
     model_name = settings.llm_model or settings.gemini_model
-    # Strip LiteLLM provider prefix (e.g. "vertex_ai/gemini-2.5-flash" -> "gemini-2.5-flash")
+    # Strip LiteLLM provider prefix (e.g. "vertex_ai/gemini-3.5-flash" -> "gemini-3.5-flash")
     if "/" in model_name:
         model_name = model_name.split("/", 1)[1]
     retry_opts = http_retry_options_from_settings(settings)

tests/test_llm_provider.py

@@ -20,6 +20,11 @@ def test_default_provider_is_gemini():
     assert s.llm_provider == "gemini"
 
 
+def test_default_gemini_model():
+    s = Settings(google_api_key="test-key")
+    assert s.gemini_model == "gemini-3.5-flash"
+
+
 def test_llm_model_defaults_to_none():
     s = Settings(google_api_key="test-key")
     assert s.llm_model is None