diff --git a/cloud/deploy/.dockerignore b/cloud/deploy/.dockerignore
new file mode 100644
index 0000000..d96e442
--- /dev/null
+++ b/cloud/deploy/.dockerignore
@@ -0,0 +1,14 @@
+# Keep the build context small — the images sync from pyproject + uv.lock, not these.
+.git
+.venv
+**/__pycache__
+**/*.pyc
+**/node_modules
+apps/web/.next
+apps/vscode/dist
+website
+**/.pytest_cache
+**/.mypy_cache
+**/.ruff_cache
+*.db
+.volo
diff --git a/cloud/deploy/Dockerfile.api b/cloud/deploy/Dockerfile.api
new file mode 100644
index 0000000..11b538a
--- /dev/null
+++ b/cloud/deploy/Dockerfile.api
@@ -0,0 +1,26 @@
+# Volo Cloud — API image (the FastAPI control plane).
+# Build context is the REPO ROOT (the app is a uv-workspace member):
+#   docker build -f cloud/deploy/Dockerfile.api -t volo-cloud-api .
+#
+# Syncs the whole workspace so transitive workspace deps (volo-core, volo-api for the auth seam,
+# volo-runner, …) and the `serve` extra (uvicorn) are present. Optimizing to a slim, cloud-only
+# image is a follow-up.
+FROM python:3.12-slim
+
+# uv provides fast, reproducible installs from the committed uv.lock.
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
+
+ENV UV_COMPILE_BYTECODE=1 \
+    UV_LINK_MODE=copy \
+    PYTHONUNBUFFERED=1 \
+    # deny anonymous callers in a hosted deployment (M30 auth gate)
+    VOLO_REQUIRE_AUTH=true
+
+WORKDIR /app
+COPY . .
+RUN uv sync --frozen --all-packages --all-extras
+
+EXPOSE 8080
+# uvicorn calls the app factory create_cloud_app()
+CMD ["uv", "run", "uvicorn", "volo_cloud.app:create_cloud_app", \
+     "--factory", "--host", "0.0.0.0", "--port", "8080"]
diff --git a/cloud/deploy/Dockerfile.worker b/cloud/deploy/Dockerfile.worker
new file mode 100644
index 0000000..a20371c
--- /dev/null
+++ b/cloud/deploy/Dockerfile.worker
@@ -0,0 +1,22 @@
+# Volo Cloud — sim-minutes worker image (M27).
+# Build context is the REPO ROOT:
+#   docker build -f cloud/deploy/Dockerfile.worker -t volo-cloud-worker .
+#
+# Runs the queue drain loop: claim a queued SimJob, run the reliability suite, meter wall-clock as
+# sim-minutes, charge the workspace quota, store the report. Point it at the same VOLO_DB_URL as the
+# API. A job's agent is only executed if it is in VOLO_SIM_AGENT_ALLOWLIST (safe-by-default,
+# ADR-0033) — production should additionally sandbox this container per job.
+FROM python:3.12-slim
+
+COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
+
+ENV UV_COMPILE_BYTECODE=1 \
+    UV_LINK_MODE=copy \
+    PYTHONUNBUFFERED=1
+
+WORKDIR /app
+COPY . .
+RUN uv sync --frozen --all-packages --all-extras
+
+# --poll: seconds between queue polls when idle. Override with --once to drain and exit.
+CMD ["uv", "run", "volo-cloud-worker", "--poll", "2"]
diff --git a/cloud/deploy/README.md b/cloud/deploy/README.md
new file mode 100644
index 0000000..635e7d3
--- /dev/null
+++ b/cloud/deploy/README.md
@@ -0,0 +1,50 @@
+# Deploying Volo Cloud (hosted sim-minutes)
+
+The commercial control plane (M26/M30) + the sim-minutes worker (M27) run locally on SQLite with
+zero infra. This directory packages them for a hosted deployment: two container images (API +
+worker) and a Fly.io manifest. **Nothing here changes application behavior** — it's packaging.
+
+## Images
+
+Both build from the **repo root** (the cloud app is a uv-workspace member):
+
+```bash
+docker build -f cloud/deploy/Dockerfile.api    -t volo-cloud-api    .
+docker build -f cloud/deploy/Dockerfile.worker -t volo-cloud-worker .
+```
+
+- **API** — `uvicorn volo_cloud.app:create_cloud_app --factory` on port 8080.
+- **Worker** — `volo-cloud-worker --poll 2`, the queue drain loop.
+
+## Configuration (env / secrets)
+
+| Variable | Purpose |
+|---|---|
+| `VOLO_DB_URL` | **Postgres** DSN shared by API + worker (e.g. `postgresql://…`). Defaults to local SQLite if unset. |
+| `VOLO_REQUIRE_AUTH` | `true` in a hosted deployment — denies anonymous callers (set in the image/manifest). |
+| `VOLO_JWT_JWKS` *or* `VOLO_JWT_SECRET` | SSO: an IdP's JWKS (RS256) or a shared secret (HS256). `VOLO_JWT_ISS` / `VOLO_JWT_AUD` optional. |
+| `VOLO_SIM_AGENT_ALLOWLIST` | Comma-separated agents the **worker** may execute (safe-by-default, ADR-0033). |
+
+> **Security:** the worker executes a job's agent code only if it's allowlisted. In production also
+> sandbox the worker container per job — treat it as an untrusted-code boundary.
+
+## Fly.io
+
+`fly.toml` runs one image with two processes (`app`, `worker`). A first deploy:
+
+```bash
+fly launch --copy-config --no-deploy          # uses cloud/deploy/fly.toml
+fly postgres create && fly postgres attach …  # sets VOLO_DB_URL (DATABASE_URL) as a secret
+fly secrets set VOLO_JWT_JWKS='{"keys":[…]}' VOLO_SIM_AGENT_ALLOWLIST='my_pkg.agent:run'
+fly deploy
+```
+
+Scale the worker independently of the API:
+
+```bash
+fly scale count app=1 worker=2
+```
+
+The same images run on any container host (Cloud Run, ECS, a plain VM) — set the same env and run
+the two commands. A slimmer cloud-only image (syncing just `volo-cloud` instead of the whole
+workspace) is a reasonable follow-up optimization.
diff --git a/cloud/deploy/fly.toml b/cloud/deploy/fly.toml
new file mode 100644
index 0000000..3ed0a7d
--- /dev/null
+++ b/cloud/deploy/fly.toml
@@ -0,0 +1,29 @@
+# Fly.io manifest for the Volo Cloud control plane (API + sim-minutes worker).
+# One image, two processes. Set secrets (DB, auth, allowlist) with `fly secrets set` — see README.
+app = "volo-cloud"
+primary_region = "iad"
+
+[build]
+  dockerfile = "cloud/deploy/Dockerfile.api"
+
+[env]
+  # Deny anonymous callers; a real IdP is wired via VOLO_JWT_JWKS / VOLO_JWT_SECRET (secrets).
+  VOLO_REQUIRE_AUTH = "true"
+
+# The API serves HTTP; the worker drains the sim-minutes queue. Both run from the same image;
+# the worker process overrides the command.
+[processes]
+  app = "uv run uvicorn volo_cloud.app:create_cloud_app --factory --host 0.0.0.0 --port 8080"
+  worker = "uv run volo-cloud-worker --poll 2"
+
+[http_service]
+  internal_port = 8080
+  force_https = true
+  auto_stop_machines = "stop"
+  auto_start_machines = true
+  min_machines_running = 0
+  processes = ["app"]
+
+[[vm]]
+  size = "shared-cpu-1x"
+  memory = "512mb"