From 97d0c7a8ef1267af80a450246dd82f00fa299068 Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Sun, 9 Aug 2026 16:53:14 -0700
Subject: [PATCH] [eric] apps: an app nobody is editing serves its built bundle
 with no dev-server process (~270MB -> ~0), vite reboots the moment an agent
 binds or the dist goes stale

---
 backend/apps/agents/manager/AgentLaunch.py    |   7 ++
 backend/apps/outputs/outputs.py               |  17 ++-
 backend/apps/outputs/runtime.py               |  39 +++++-
 backend/apps/outputs/static_serve.py          | 112 ++++++++++++++++++
 .../webapp_template/frontend/vite.config.ts   |   2 +
 backend/tests/test_serve_static_mode.py       | 106 +++++++++++++++++
 6 files changed, 277 insertions(+), 6 deletions(-)
 create mode 100644 backend/apps/outputs/static_serve.py
 create mode 100644 backend/tests/test_serve_static_mode.py

diff --git a/backend/apps/agents/manager/AgentLaunch.py b/backend/apps/agents/manager/AgentLaunch.py
index 179110ae..3b544556 100644
--- a/backend/apps/agents/manager/AgentLaunch.py
+++ b/backend/apps/agents/manager/AgentLaunch.py
@@ -61,6 +61,13 @@ class AgentLaunch(AgentManagerProtocol):
         ):
             from backend.apps.outputs.workspace_io import app_workspace_dir
             bound = app_workspace_dir(config.selected_app_output_ids[0])
+            if bound:
+                # The user is about to EDIT this app: a serve-static runtime must boot vite now or the agent's changes render nowhere (ENG-209).
+                try:
+                    from backend.apps.outputs.runtime import manager as p_rt_manager
+                    await p_rt_manager.ensure_editing(bound)
+                except Exception:
+                    pass
             if bound:
                 config.target_directory = bound
 
diff --git a/backend/apps/outputs/outputs.py b/backend/apps/outputs/outputs.py
index fa38e207..6af4b347 100644
--- a/backend/apps/outputs/outputs.py
+++ b/backend/apps/outputs/outputs.py
@@ -120,18 +120,23 @@ async def serve_workspace_file(workspace_id: str, filepath: str, p_d: str = ""):
     if not os.path.isfile(full_path):
         raise HTTPException(status_code=404, detail="File not found")
 
-    with open(full_path) as f:
-        content = f.read()
-
-    if filepath == "index.html":
+    # endswith, not equality: serve-mode delivers frontend/dist/index.html through this same route and needs the identical injection + token rewrite (ENG-209).
+    if filepath.endswith("index.html"):
+        with open(full_path) as f:
+            content = f.read()
         input_json, result_json = decode_data_param(p_d) if p_d else ("{}", "null")
         backend_url_json = backend_url_for_workspace(workspace_id)
         content = inject_data_into_html(content, input_json, result_json, backend_url_json, with_runtime=True)
         # Iframe sub-resource fetches (<link>, <script src>, <img>) drop the parent's ?token= query string, so rewrite the HTML to put the token back on every relative URL; otherwise sub-resources 401.
         content = inject_token_into_relative_urls(content, get_auth_token())
+        mime, _ = mimetypes.guess_type(filepath)
+        return Response(content=content, media_type=mime or "text/plain")
 
+    # Binary-safe for everything else: a built bundle carries fonts and images that a text read would corrupt.
+    with open(full_path, "rb") as f:
+        raw = f.read()
     mime, _ = mimetypes.guess_type(filepath)
-    return Response(content=content, media_type=mime or "text/plain")
+    return Response(content=raw, media_type=mime or "application/octet-stream")
 
 
 @outputs.router.get("/{output_id}/serve/{filepath:path}")
@@ -460,6 +465,8 @@ def runtime_status_payload(workspace_id: str, instance: int = 1) -> dict:
         "running": rt.running,
         # 'spawned' vs 'serving': ready flips only once the primary port answered the bind poll (and un-flips when the process dies or is frozen).
         "ready": rt.ready,
+        # True when the app is served as a built bundle with no dev-server process (ENG-209).
+        "serve_static": rt.serve_static,
         "port": rt.port,
         "serving_url": serving_url,
         "has_backend_file": rt.has_backend_file,
diff --git a/backend/apps/outputs/runtime.py b/backend/apps/outputs/runtime.py
index 201bb23e..a7b3eca6 100644
--- a/backend/apps/outputs/runtime.py
+++ b/backend/apps/outputs/runtime.py
@@ -108,6 +108,8 @@ class AppRuntime:
         # Old-mode: `port` is the backend.py port. New-mode: `port` is the workspace's optional FastAPI backend (only set if BACKEND_PORT!=NONE) and `frontend_port` is the Vite dev server port. Both Nones until start() decides what's there.
         self.port: Optional[int] = None
         self.frontend_port: Optional[int] = None
+        # Serve-mode (ENG-209): a fresh built bundle + no agent editing means NO process at all; the workspace serve route delivers frontend/dist and this flag makes ready/frontend_url say so.
+        self.serve_static: bool = False
         # New-mode only: flips True once something is actually listening on frontend_port (we kick off a background poll task in p_start_new_mode). frontend_url returns null until this flips, so the preview pane doesn't try to navigate to an unbound port and show a "Site can't be reached" error mid-npm-install.
         self.p_frontend_ready: bool = False
         # True while the process tree is SIGSTOP'd in the idle pool. A frozen vite still holds its port but can't answer it, so frontend_url must stay null while suspended (else the webview loads a dead port = the ERR_FAILED on fast app-switching).
@@ -174,6 +176,8 @@ class AppRuntime:
         """True only when the runtime is actually SERVING (process alive, not frozen, and its
         primary port answered the bind poll), so callers can tell 'spawned' from 'serving'.
         Old-mode workspaces have no bind poll; a live process is their best readiness signal."""
+        if self.serve_static:
+            return True
         if not self.running or self.p_suspended:
             return False
         if self.is_new_mode:
@@ -183,6 +187,11 @@ class AppRuntime:
     @property
     def frontend_url(self) -> Optional[str]:
         # Gated on `_frontend_ready` (set by the background bind-poll task in p_start_new_mode) so the preview pane only switches over once Vite is actually accepting connections. Without this, the editor flashes a "Site can't be reached" error while `npm install` is running. Also gated on `running`: a vite that crashed or got orphaned still has _frontend_ready=True, and handing the webview that dead port is the ERR_FAILED you see on reopen. No live process, no URL. And gated on `not _suspended`: a SIGSTOP'd idle runtime is "running" (returncode is None) but frozen, so its port won't answer.
+        if self.serve_static:
+            # The existing workspace serve route: index injection + token rewrite apply, and the dist's relative assets resolve under the same path.
+            from backend.auth import init_auth_token
+            p_port = os.environ.get("OPENSWARM_PORT", "8324")
+            return f"http://127.0.0.1:{p_port}/api/outputs/workspace/{self.workspace_id}/serve/frontend/dist/index.html?token={init_auth_token()}"
         if self.frontend_port and self.p_frontend_ready and self.running and not self.p_suspended:
             return f"http://127.0.0.1:{self.frontend_port}/"
         return None
@@ -230,6 +239,14 @@ class AppRuntime:
             return await self.p_start_old_mode()
 
     async def p_start_new_mode(self) -> bool:
+        # Serve-mode (ENG-209): a fresh built bundle + nobody editing = no process at all. Primary
+        # instance only (secondaries are explicitly "another independent window", keep them live).
+        if self.instance == 1:
+            from backend.apps.outputs.static_serve import static_fresh, workspace_being_edited
+            if static_fresh(self.workspace_path) and not workspace_being_edited(self.workspace_path):
+                self.serve_static = True
+                self.p_broadcast(LogLine("runtime", "[runtime] serving the built bundle (no dev server); editing an app boots vite automatically"))
+                return True
         # Legacy workspaces (scaffolded pre-multi-instance) ignore the forced ports; self-heal their run.sh so a second instance stops colliding on the primary's ports.
         ensure_force_port_shim(self.workspace_path)
         env_path = os.path.join(self.workspace_path, ".env")
@@ -672,8 +689,15 @@ class AppRuntimeManager:
                 # Workspace paths shouldn't change for a given id, but if somehow they did (e.g. the user moved the workspace folder), trust the latest caller; they have the current truth.
                 rt.workspace_path = workspace_path
             self.p_attached[key] = self.p_attached.get(key, 0) + 1
-        if not revived and not rt.running:
+        if not revived and not rt.running and not rt.serve_static:
             await rt.start()
+        # A serve-static runtime re-checks its world on every attach: an agent may have bound to the
+        # workspace since, or the dist may have gone stale; either flips it back to a real vite boot.
+        if rt.serve_static:
+            from backend.apps.outputs.static_serve import static_fresh, workspace_being_edited
+            if workspace_being_edited(rt.workspace_path) or not static_fresh(rt.workspace_path):
+                rt.serve_static = False
+                await rt.start()
         # Stop any dead idle runtime outside the lock to avoid blocking.
         if dead is not None:
             try:
@@ -682,6 +706,15 @@ class AppRuntimeManager:
                 logger.exception("failed to reap dead idle runtime %s", key)
         return rt
 
+    async def ensure_editing(self, workspace_path: str) -> None:
+        """An agent just bound to this workspace: any serve-static runtime for it must become a real
+        vite dev server, or the agent edits files the user never sees."""
+        norm = os.path.realpath(workspace_path)
+        for rt in list(self.runtimes.values()) + list(self.idle_lru.values()):
+            if os.path.realpath(rt.workspace_path) == norm and rt.serve_static:
+                rt.serve_static = False
+                await rt.start()
+
     async def detach(self, workspace_id: str, instance: int = 1) -> None:
         key = runtime_key(workspace_id, instance)
         to_idle: Optional[AppRuntime] = None
@@ -695,6 +728,10 @@ class AppRuntimeManager:
             rt = self.runtimes.pop(key, None)
             if rt is None:
                 return
+            # Park-time dist build (ENG-209): a vite runtime going idle is the moment to bake the static bundle, so the NEXT open serves it processlessly. Fire-and-forget; failure just means vite again.
+            if rt.running and not rt.serve_static and rt.instance == 1 and is_new_mode(rt.workspace_path):
+                from backend.apps.outputs.static_serve import build_dist_in_background
+                asyncio.create_task(build_dist_in_background(rt.workspace_path, rt.workspace_id))
             # If the process is already dead, no point keeping it around; just clean up. Otherwise move to the LRU AND SIGSTOP the process tree so it consumes 0% CPU while idle. The matching SIGCONT lives in attach() above.
             if not rt.running:
                 to_reap.append(rt)
diff --git a/backend/apps/outputs/static_serve.py b/backend/apps/outputs/static_serve.py
new file mode 100644
index 00000000..46bcec19
--- /dev/null
+++ b/backend/apps/outputs/static_serve.py
@@ -0,0 +1,112 @@
+"""Serve-mode support (ENG-209): an open app that nobody is editing costs a whole vite dev server
+(~270MB). When a FRESH built bundle exists and no agent is bound to the workspace, the runtime
+serves `frontend/dist` statically through the existing workspace serve route instead of spawning
+anything; vite comes back the moment an agent starts editing."""
+
+import asyncio
+import logging
+import os
+import subprocess
+from typing import Optional
+
+from typeguard import typechecked
+
+logger = logging.getLogger(__name__)
+
+
+@typechecked
+def dist_index(workspace_path: str) -> str:
+    return os.path.join(workspace_path, "frontend", "dist", "index.html")
+
+
+@typechecked
+def p_newest_source_mtime(workspace_path: str) -> float:
+    newest = 0.0
+    fe = os.path.join(workspace_path, "frontend")
+    for probe in ("package.json", "vite.config.ts", "index.html"):
+        p = os.path.join(fe, probe)
+        if os.path.isfile(p):
+            newest = max(newest, os.path.getmtime(p))
+    src = os.path.join(fe, "src")
+    for dirpath, dirnames, filenames in os.walk(src):
+        dirnames[:] = [d for d in dirnames if d != "node_modules"]
+        for f in filenames:
+            try:
+                newest = max(newest, os.path.getmtime(os.path.join(dirpath, f)))
+            except OSError:
+                pass
+    return newest
+
+
+@typechecked
+def static_fresh(workspace_path: str) -> bool:
+    """A dist we can honestly serve: exists, newer than every source file, and built with RELATIVE
+    asset paths (vite `base: './'`). Absolute /assets/ URLs would resolve against the backend host
+    root and 404, so a dist built by an older template is refused rather than served broken."""
+    idx = dist_index(workspace_path)
+    if not os.path.isfile(idx):
+        return False
+    try:
+        with open(idx, encoding="utf-8", errors="ignore") as f:
+            html = f.read()
+    except OSError:
+        return False
+    if 'src="/' in html or 'href="/' in html:
+        return False
+    return os.path.getmtime(idx) >= p_newest_source_mtime(workspace_path)
+
+
+@typechecked
+def workspace_being_edited(workspace_path: str) -> bool:
+    """True when a live agent session is bound to this workspace (launch sets the chat's cwd to the
+    app it edits), so serve-mode never hides an agent's in-flight changes behind a stale bundle."""
+    try:
+        from backend.apps.agents.agents import agent_manager
+        norm = os.path.realpath(workspace_path)
+        for session in agent_manager.sessions.values():
+            if session.status not in ("running", "waiting_approval"):
+                continue
+            cwd = getattr(session, "cwd", None)
+            if cwd and os.path.realpath(cwd) == norm:
+                return True
+    except Exception:
+        logger.debug("being-edited probe failed; assuming edited (vite)", exc_info=True)
+        return True
+    return False
+
+
+@typechecked
+async def build_dist_in_background(workspace_path: str, workspace_id: str) -> Optional[int]:
+    """Fire `npm run build` for a parked app so the NEXT open can serve static. Best-effort: any
+    missing precondition (no node_modules, no build script) just means vite next time. Returns the
+    exit code, or None when the build was skipped."""
+    fe = os.path.join(workspace_path, "frontend")
+    if not os.path.isdir(os.path.join(fe, "node_modules")):
+        return None
+    if static_fresh(workspace_path):
+        return None
+    node_bin = os.path.dirname(os.environ.get("OPENSWARM_NODE_PATH", "") or "")
+    env = {**os.environ}
+    if node_bin:
+        env["PATH"] = f"{node_bin}:{env.get('PATH', '')}"
+    log_path = os.path.join(workspace_path, ".openswarm", "build.log")
+    os.makedirs(os.path.dirname(log_path), exist_ok=True)
+    try:
+        with open(log_path, "w", encoding="utf-8") as log:
+            proc = await asyncio.create_subprocess_exec(
+                "npm", "run", "build",
+                cwd=fe, env=env, stdout=log, stderr=subprocess.STDOUT,
+            )
+            code = await asyncio.wait_for(proc.wait(), timeout=180)
+        logger.info("background dist build for %s exited %s", workspace_id, code)
+        return code
+    except asyncio.TimeoutError:
+        try:
+            proc.kill()
+        except Exception:
+            pass
+        logger.warning("background dist build for %s timed out", workspace_id)
+        return -1
+    except Exception:
+        logger.debug("background dist build for %s failed to spawn", workspace_id, exc_info=True)
+        return None
diff --git a/backend/apps/outputs/webapp_template/frontend/vite.config.ts b/backend/apps/outputs/webapp_template/frontend/vite.config.ts
index b8ffffcb..f079a0f6 100644
--- a/backend/apps/outputs/webapp_template/frontend/vite.config.ts
+++ b/backend/apps/outputs/webapp_template/frontend/vite.config.ts
@@ -49,6 +49,8 @@ export default defineConfig(({ mode }) => {
   const backendEnabled = backendPort && backendPort !== 'NONE';
 
   return {
+    // Relative asset URLs: the built bundle is served under /api/outputs/workspace/<id>/serve/frontend/dist/, where absolute /assets/ paths would 404 (ENG-209 serve-mode).
+    base: './',
     cacheDir: sharedViteCacheDir(),
     plugins: [
       react(),
diff --git a/backend/tests/test_serve_static_mode.py b/backend/tests/test_serve_static_mode.py
new file mode 100644
index 00000000..1a7bd1df
--- /dev/null
+++ b/backend/tests/test_serve_static_mode.py
@@ -0,0 +1,106 @@
+"""ENG-209: an open app nobody is editing serves its built bundle with NO process; vite comes back
+the moment an agent binds or the dist goes stale. Every transition is covered, since a wrong switch
+would break preview for every app."""
+
+import asyncio
+import os
+import time
+
+from backend.apps.outputs import static_serve
+from backend.apps.outputs.runtime import AppRuntime, AppRuntimeManager
+
+
+def p_seed(tmp_path, relative_assets=True, with_dist=True):
+    ws = tmp_path / "ws"
+    fe = ws / "frontend"
+    (fe / "src").mkdir(parents=True)
+    (fe / "src" / "App.tsx").write_text("export {}")
+    (fe / "package.json").write_text("{}")
+    (ws / "run.sh").write_text("#!/bin/bash\n")
+    if with_dist:
+        dist = fe / "dist"
+        dist.mkdir()
+        src = "./assets/index-abc.js" if relative_assets else "/assets/index-abc.js"
+        (dist / "index.html").write_text(f'<html><script src="{src}"></script></html>')
+    return str(ws)
+
+
+def test_static_fresh_requires_dist_newer_than_source_and_relative_assets(tmp_path):
+    ws = p_seed(tmp_path)
+    assert static_serve.static_fresh(ws) is True
+    time.sleep(0.02)
+    os.utime(os.path.join(ws, "frontend", "src", "App.tsx"))
+    assert static_serve.static_fresh(ws) is False
+    assert static_serve.static_fresh(p_seed(tmp_path / "abs", relative_assets=False)) is False
+    assert static_serve.static_fresh(p_seed(tmp_path / "nodist", with_dist=False)) is False
+
+
+def test_start_serves_static_when_not_edited(tmp_path, monkeypatch):
+    ws = p_seed(tmp_path)
+    monkeypatch.setattr(static_serve, "workspace_being_edited", lambda p: False)
+    rt = AppRuntime("ws-t", ws)
+    assert asyncio.run(rt.start()) is True
+    assert rt.serve_static is True and rt.process is None and rt.ready is True
+    assert "/serve/frontend/dist/index.html" in (rt.frontend_url or "")
+
+
+def test_start_skips_serve_when_edited(tmp_path, monkeypatch):
+    ws = p_seed(tmp_path)
+    monkeypatch.setattr(static_serve, "workspace_being_edited", lambda p: True)
+    rt = AppRuntime("ws-t2", ws)
+    spawned = {"n": 0}
+
+    async def p_no_spawn(env):
+        spawned["n"] += 1
+        return None, ws, "stub"
+    monkeypatch.setattr(rt, "p_resolve_launch", p_no_spawn)
+    try:
+        asyncio.run(rt.start())
+    except Exception:
+        pass
+    assert rt.serve_static is False and spawned["n"] == 1
+
+
+def p_manager_no_watcher(monkeypatch):
+    # The restart-sentinel watcher is a forever loop; tests must not arm it or asyncio.run leaks a task.
+    monkeypatch.setattr(AppRuntimeManager, "p_ensure_restart_watcher", lambda self: None)
+    return AppRuntimeManager()
+
+
+def test_ensure_editing_flips_serve_runtime_to_vite(tmp_path, monkeypatch):
+    ws = p_seed(tmp_path)
+    mgr = p_manager_no_watcher(monkeypatch)
+    rt = AppRuntime("ws-t", ws)
+    rt.serve_static = True
+    restarted = {"n": 0}
+
+    async def p_fake_start():
+        restarted["n"] += 1
+        return True
+    monkeypatch.setattr(rt, "start", p_fake_start)
+    mgr.runtimes["ws-t"] = rt
+    asyncio.run(mgr.ensure_editing(ws))
+    assert rt.serve_static is False and restarted["n"] == 1
+
+
+def test_attach_recheck_reboots_when_dist_goes_stale(tmp_path, monkeypatch):
+    ws = p_seed(tmp_path)
+    mgr = p_manager_no_watcher(monkeypatch)
+    rt = AppRuntime("ws-t", ws)
+    rt.serve_static = True
+    restarted = {"n": 0}
+
+    async def p_fake_start():
+        restarted["n"] += 1
+        return True
+    monkeypatch.setattr(rt, "start", p_fake_start)
+    mgr.runtimes["ws-t"] = rt
+
+    monkeypatch.setattr(static_serve, "workspace_being_edited", lambda p: False)
+    asyncio.run(mgr.attach("ws-t", ws))
+    assert rt.serve_static is True and restarted["n"] == 0
+
+    time.sleep(0.02)
+    os.utime(os.path.join(ws, "frontend", "src", "App.tsx"))
+    asyncio.run(mgr.attach("ws-t", ws))
+    assert rt.serve_static is False and restarted["n"] == 1