chore(agent): upgrade agent default model to gemini-3.6-flash

echo/agent/settings.py

@@ -11,7 +11,7 @@ class Settings(BaseSettings):
     echo_api_url: str = Field(default="http://localhost:8000/api", alias="ECHO_API_URL")
     # Vertex model id (no provider prefix). Gemini 3.x is served on the
     # global Vertex host only; see vertex_api_endpoint below.
-    llm_model: str = Field(default="gemini-3.5-flash", alias="LLM_MODEL")
+    llm_model: str = Field(default="gemini-3.6-flash", alias="LLM_MODEL")
     vertex_location: str = Field(default="eu", alias="VERTEX_LOCATION")
     # Pinning the global host while keeping locations/<region> in the request
     # path mirrors the server's LiteLLM config (validated in production):

echo/agent/tests/test_agent_graph.py

@@ -159,7 +159,7 @@ def from_service_account_info(info, scopes=None):
 def test_build_llm_prefers_explicit_vertex_credentials(monkeypatch):
     get_settings.cache_clear()
     fake_chat, fake_creds = _fake_vertex_chat(monkeypatch)
-    monkeypatch.setenv("LLM_MODEL", "gemini-3.5-flash")
+    monkeypatch.setenv("LLM_MODEL", "gemini-3.6-flash")
     monkeypatch.setenv("VERTEX_PROJECT", "vertex-project")
     monkeypatch.setenv("VERTEX_LOCATION", "europe-west4")
     monkeypatch.setenv("VERTEX_CREDENTIALS", '{"type":"service_account","project_id":"explicit"}')
@@ -168,7 +168,7 @@ def test_build_llm_prefers_explicit_vertex_credentials(monkeypatch):
     llm = _build_llm()
 
     assert isinstance(llm, fake_chat)
-    assert llm.kwargs["model_name"] == "gemini-3.5-flash"
+    assert llm.kwargs["model_name"] == "gemini-3.6-flash"
     assert llm.kwargs["project"] == "vertex-project"
     assert llm.kwargs["location"] == "europe-west4"
     assert isinstance(llm.kwargs["credentials"], fake_creds)
@@ -198,7 +198,7 @@ def test_build_llm_falls_back_to_service_account_project_id(monkeypatch):
     fake_chat, fake_creds = _fake_vertex_chat(monkeypatch)
     monkeypatch.delenv("VERTEX_CREDENTIALS", raising=False)
     monkeypatch.delenv("VERTEX_PROJECT", raising=False)
-    monkeypatch.setenv("LLM_MODEL", "gemini-3.5-flash")
+    monkeypatch.setenv("LLM_MODEL", "gemini-3.6-flash")
     monkeypatch.setenv("VERTEX_LOCATION", "eu")
     monkeypatch.setenv("GCP_SA_JSON", '{"type":"service_account","project_id":"sa-project"}')
 

echo/docs/agentic-chat-design.md

@@ -11,7 +11,7 @@ agent to specific conversations (today's deep dive collapses into a scoped agent
 mode). The agent gains a per-workspace virtual file system containing the product
 docs and meta-skills, acts within the logged-in user's auth boundary (including
 writes like project settings and portal editor), auto-onboards new users, and serves
-as primary support. Model: gemini-3.5-flash.
+as primary support. Model: gemini-3.6-flash.
 
 ## Decisions taken
 
@@ -22,7 +22,7 @@ as primary support. Model: gemini-3.5-flash.
    CONFLICTING). Worth taking: server-side grep with chunk-level citations, citation
    rendering, agentic title generation, test suites. Not taking: the raw Vertex
    Anthropic model swap.
-3. Agent model goes through the LiteLLM router config (gemini-3.5-flash), replacing
+3. Agent model goes through the LiteLLM router config (gemini-3.6-flash), replacing
    both main's raw GEMINI_API_KEY client and the PR's raw Vertex Anthropic client.
    If a stronger reviewer tier is added later, the cheap model never decides to
    escalate on its own (scheduled or user-explicit only).

echo/frontend/src/config.ts

@@ -177,7 +177,7 @@ export const ENABLE_ANNOUNCEMENTS = true;
 export const ENABLE_DISPLAY_CONVERSATION_LINKS = true;
 export const ENABLE_WEBHOOKS = true;
 // On everywhere except production: local, testing/dev, next/staging.
-// Re-enabled 2026-07-02 with the agent on gemini-3.5-flash via Vertex and the
+// Re-enabled 2026-07-02 with the agent on gemini-3.6-flash via Vertex and the
 // #573 harvest (server-side grep, chunk citations, titles) on main.
 export const ENABLE_AGENTIC_CHAT = byEnv({ production: false }, true);
 // Re-enabled on echo-next (2026-06-21). Runtime render gate only; the build