studio: agent window, a chat with a model that edits the scene

✦ Agent in the Studio's top bar opens a floating chat window. The model
(any OpenAI-compatible endpoint with tool calling; default B4L's inference
server, https://inference.p0.contact/v1) gets the 15 MCP tools plus two page
tools (select_object, set_view) and the Studio's state (selection, level,
rotation) in its system prompt. Answers and reasoning stream in; tool calls
show as rows with arguments and results; object ids link to the object;
Stop ends a turn; Undo takes the agent's edits back. The window moves by
its title bar and resizes at its top-left corner.

- studio/agent.js: shared by both front ends (waits for studio.js's
  isokit:studio event); MCP over /mcp or isokit.mcp in the web Studio
- isokit studio: /api/agent/{config,models,chat} relay the calls on their
  own threads (ureq, chunked and flushed per read), so the key stays on the
  machine and the endpoint needs no CORS; settings from ISOKIT_LLM_URL /
  _KEY / _MODEL or agent.json in the user's config folder
- web Studio: calls the endpoint from the page with a key kept in the
  browser; build_web.py --agent-url names a same-site relay as default; the
  deploy builds with /llm/v1, which deploy/Caddyfile forwards to the
  inference server (its nginx refuses the CORS preflight)
- tools/mock_llm.py + tools/agent_checks.py: the agent window checks in both
  check_studio.py and check_web.py

tools/check_web.py

@@ -7,8 +7,9 @@ Serves the build on a local port and checks: the demo loads from the bundled
 files, placing / dragging / props / undo, persistence across reloads, the code
 editor (script error and repair), new / export / import (.zip, stored and
 deflated), page build and preview, MCP in the page and WebMCP, two tabs on one
-project, and a folder project (the origin-private file system stands in for a
-folder picked with showDirectoryPicker: same handle API) following outside edits.
+project, a folder project (the origin-private file system stands in for a
+folder picked with showDirectoryPicker: same handle API) following outside
+edits, and the agent window on a mock model (tools/mock_llm.py, started here).
 """
 import io
 import json
@@ -20,6 +21,8 @@ from pathlib import Path
 
 from playwright.sync_api import sync_playwright
 
+import agent_checks
+
 ROOT = Path(__file__).resolve().parent.parent
 OUT = Path(sys.argv[1]) if len(sys.argv) > 1 else ROOT / "dist" / "web-shots"
 DIST = Path(sys.argv[2]) if len(sys.argv) > 2 else ROOT / "dist" / "web"
@@ -60,6 +63,7 @@ WEBMCP_STUB = "navigator.modelContext = { provideContext: function (c) { window.
 
 srv = subprocess.Popen([sys.executable, "-m", "http.server", str(PORT), "--bind", "127.0.0.1", "--directory", str(DIST)],
                        stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+mock = agent_checks.start_mock()
 time.sleep(1.0)
 try:
     with sync_playwright() as p:
@@ -186,6 +190,12 @@ try:
             return t.execute({id: "agent-lamp", tag: "via WebMCP"}); }""")
         check(r["isError"] is False and obj(pg, "agent-lamp")["tag"] == "via WebMCP", "a WebMCP tool call edits the scene")
 
+        # ---------------------------------------------------------------- agent window (mock model, key kept in the browser)
+        pg.evaluate("(u) => localStorage.setItem('isokit-agent', JSON.stringify({url: u, key: 'test-key'}))", agent_checks.MOCK + "/v1")
+        agent_checks.run(pg, lambda: doc(pg)["objects"], lambda n: shot(pg, "web-" + n), check, can_set_key=True)
+        stored = pg.evaluate("() => JSON.parse(localStorage.getItem('isokit-agent'))")
+        check(stored.get("key") == "test-key" and not stored.get("model"), f"settings stay in this browser {stored}")
+
         # ---------------------------------------------------------------- two tabs on one project
         pg2 = ctx.new_page()
         watch(pg2)
@@ -309,10 +319,13 @@ try:
         pg.wait_for_timeout(400)
         shot(pg, "web-narrow.jpg")
 
+        # the browser logs the 401 of the agent window's wrong-key check
+        errs = [e for e in errs if "status of 401" not in e]
         check(not errs, f"no page errors {errs[:5]}")
         b.close()
 finally:
     srv.terminate()
+    mock.terminate()
 
 print("\n%d failure(s)" % len(fails))
 sys.exit(1 if fails else 0)