diff --git a/.github/workflows/nightly-e2e.yaml b/.github/workflows/nightly-e2e.yaml
index c15f11914..1ee127492 100644
--- a/.github/workflows/nightly-e2e.yaml
+++ b/.github/workflows/nightly-e2e.yaml
@@ -162,16 +162,13 @@ jobs:
           if-no-files-found: ignore
 
   # ── GPU E2E (Ollama local inference) ──────────────────────────
-  # Enable by setting repository variable GPU_E2E_ENABLED=true
-  # (Settings → Secrets and variables → Actions → Variables)
-  #
-  # Runner labels: using 'self-hosted' for now. Refine to
-  # [self-hosted, linux, x64, gpu] once NVIDIA runner labels are confirmed.
+  # Runs on an ephemeral Brev GPU instance with Ollama pre-installed.
   gpu-e2e:
     if: github.repository == 'NVIDIA/NemoClaw' && vars.GPU_E2E_ENABLED == 'true'
-    runs-on: self-hosted
-    timeout-minutes: 60
+    runs-on: ubuntu-latest
+    timeout-minutes: 90
     env:
+      BREV_API_TOKEN: ${{ secrets.BREV_API_TOKEN }}
       NEMOCLAW_NON_INTERACTIVE: "1"
       NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE: "1"
       NEMOCLAW_SANDBOX_NAME: "e2e-gpu-ollama"
@@ -181,19 +178,59 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v6
 
-      - name: Verify GPU availability
+      - name: Install Brev CLI
+        run: |
+          curl -fsSL -o /tmp/brev.tar.gz "https://github.com/brevdev/brev-cli/releases/download/v0.6.322/brev-cli_0.6.322_linux_amd64.tar.gz"
+          sudo tar -xzf /tmp/brev.tar.gz -C /usr/local/bin brev
+          sudo chmod +x /usr/local/bin/brev
+
+      - name: Provision Brev GPU Instance & Run Test
+        env:
+          INSTANCE_NAME: e2e-gpu-nightly-${{ github.run_id }}
         run: |
-          echo "=== GPU Info ==="
-          nvidia-smi
-          echo ""
-          echo "=== VRAM ==="
-          nvidia-smi --query-gpu=name,memory.total --format=csv,noheader
-          echo ""
-          echo "=== Docker ==="
-          docker info --format '{{.ServerVersion}}'
-
-      - name: Run GPU E2E test (Ollama local inference)
-        run: bash test/e2e/test-gpu-e2e.sh
+          # Provision the GPU instance with our script. Form-created launchables could also just substitute their template ID here.
+          echo "Provisioning GPU instance..."
+          brev create --name "$INSTANCE_NAME" \
+            --flavor "t4" \
+            --startup-script "@scripts/brev-launchable-ci-gpu.sh"
+
+          echo "Waiting for readiness sentinel..."
+          export READY=0
+          for i in {1..20}; do
+             if brev exec "$INSTANCE_NAME" -- cat /var/run/nemoclaw-launchable-ready >/dev/null 2>&1; then
+               READY=1
+               break
+             fi
+             sleep 30
+          done
+
+          if [ $READY -eq 0 ]; then
+             echo "Instance did not become ready in time."
+             exit 1
+          fi
+
+          echo "Running GPU E2E tests remotely..."
+          # Pass the needed env variables when running
+          brev exec "$INSTANCE_NAME" -- bash -c \
+            "cd ~/NemoClaw && \
+             export NEMOCLAW_NON_INTERACTIVE=${NEMOCLAW_NON_INTERACTIVE} && \
+             export NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=${NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE} && \
+             export NEMOCLAW_SANDBOX_NAME=${NEMOCLAW_SANDBOX_NAME} && \
+             export NEMOCLAW_RECREATE_SANDBOX=${NEMOCLAW_RECREATE_SANDBOX} && \
+             export NEMOCLAW_PROVIDER=${NEMOCLAW_PROVIDER} && \
+             export OLLAMA_MODEL=qwen3:0.6b && \
+             bash test/e2e/test-gpu-e2e.sh"
+
+      - name: Tear down GPU instance
+        if: always()
+        run: brev delete e2e-gpu-nightly-${{ github.run_id }} || true
+
+      - name: Copy install log on failure
+        if: failure()
+        env:
+          INSTANCE_NAME: e2e-gpu-nightly-${{ github.run_id }}
+        run: |
+          brev scp "$INSTANCE_NAME":/tmp/nemoclaw-gpu-e2e-install.log /tmp/nemoclaw-gpu-e2e-install.log || true
 
       - name: Upload install log on failure
         if: failure()
@@ -203,6 +240,13 @@ jobs:
           path: /tmp/nemoclaw-gpu-e2e-install.log
           if-no-files-found: ignore
 
+      - name: Copy test log on failure
+        if: failure()
+        env:
+          INSTANCE_NAME: e2e-gpu-nightly-${{ github.run_id }}
+        run: |
+          brev scp "$INSTANCE_NAME":/tmp/nemoclaw-gpu-e2e-test.log /tmp/nemoclaw-gpu-e2e-test.log || true
+
       - name: Upload test log on failure
         if: failure()
         uses: actions/upload-artifact@v4
diff --git a/scripts/brev-launchable-ci-gpu.sh b/scripts/brev-launchable-ci-gpu.sh
new file mode 100755
index 000000000..1f0779142
--- /dev/null
+++ b/scripts/brev-launchable-ci-gpu.sh
@@ -0,0 +1,342 @@
+#!/usr/bin/env bash
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Brev launchable startup script — CI-Ready GPU
+#
+# Pre-bakes a VM with everything needed for NemoClaw E2E tests so that
+# CI runs only need to: rsync branch code → npm ci → nemoclaw onboard → test.
+#
+# What this installs:
+#   1. Docker (docker.io) — enabled and running
+#   2. Node.js 22 (nodesource)
+#   3. OpenShell CLI binary (pinned release)
+#   4. NemoClaw repo cloned with npm deps installed and TS plugin built
+#   5. Docker images pre-pulled (sandbox-base, openshell/cluster, node:22-slim)
+#   6. NVIDIA Container Toolkit (for GPU passthrough to Docker)
+#   7. Ollama + specified model pre-pulled
+#
+# What this does NOT install (intentionally):
+#   - code-server (not needed for automated CI)
+#   - VS Code themes/extensions
+#
+# Readiness detection:
+#   Writes /var/run/nemoclaw-launchable-ready when complete.
+#   Also writes "=== Ready ===" to /tmp/launch-plugin.log for backward compat.
+#
+# Usage (Brev launchable startup script — one-liner that curls this):
+#   curl -fsSL https://raw.githubusercontent.com/NVIDIA/NemoClaw/<ref>/scripts/brev-launchable-ci-gpu.sh | bash
+#
+# Environment overrides:
+#   OPENSHELL_VERSION     — OpenShell CLI release tag (default: v0.0.20)
+#   NEMOCLAW_REF          — NemoClaw git ref to clone (default: main)
+#   NEMOCLAW_CLONE_DIR    — Where to clone NemoClaw (default: ~/NemoClaw)
+#   SKIP_DOCKER_PULL      — Set to 1 to skip Docker image pre-pulls
+#   OLLAMA_MODEL          — Which model to preload (default: qwen3:0.6b, fallback qwen2.5:0.5b)
+#
+# Related:
+#   - Epic: https://github.com/NVIDIA/NemoClaw/issues/1326
+#   - Issue: https://github.com/NVIDIA/NemoClaw/issues/1327
+
+set -euo pipefail
+
+# ── Configuration ────────────────────────────────────────────────────
+OPENSHELL_VERSION="${OPENSHELL_VERSION:-v0.0.20}"
+NEMOCLAW_REF="${NEMOCLAW_REF:-main}"
+TARGET_USER="${SUDO_USER:-$(id -un)}"
+TARGET_HOME="$(getent passwd "$TARGET_USER" | cut -d: -f6)"
+NEMOCLAW_CLONE_DIR="${NEMOCLAW_CLONE_DIR:-${TARGET_HOME}/NemoClaw}"
+OLLAMA_MODEL="${OLLAMA_MODEL:-qwen3:0.6b}"
+
+LAUNCH_LOG="${LAUNCH_LOG:-/tmp/launch-plugin.log}"
+SENTINEL="/var/run/nemoclaw-launchable-ready"
+
+# Docker images to pre-pull. These are the expensive layers that cause
+# timeouts when pulled during CI runs.
+DOCKER_IMAGES=(
+  "ghcr.io/nvidia/nemoclaw/sandbox-base:latest"
+  "node:22-slim"
+)
+
+# ── Suppress apt noise ───────────────────────────────────────────────
+export DEBIAN_FRONTEND=noninteractive
+export NEEDRESTART_MODE=a
+
+# ── Logging ──────────────────────────────────────────────────────────
+mkdir -p "$(dirname "$LAUNCH_LOG")"
+exec > >(tee -a "$LAUNCH_LOG") 2>&1
+
+_ts() { date '+%H:%M:%S'; }
+info() { printf '\033[0;32m[%s ci-gpu]\033[0m %s\n' "$(_ts)" "$1"; }
+warn() { printf '\033[1;33m[%s ci-gpu]\033[0m %s\n' "$(_ts)" "$1"; }
+fail() {
+  printf '\033[0;31m[%s ci-gpu]\033[0m %s\n' "$(_ts)" "$1"
+  exit 1
+}
+
+# ── Retry helper ─────────────────────────────────────────────────────
+# Usage: retry 3 10 "description" command arg1 arg2
+retry() {
+  local max_attempts="$1" sleep_sec="$2" desc="$3"
+  shift 3
+  local attempt=1
+  while true; do
+    if "$@"; then
+      return 0
+    fi
+    if ((attempt >= max_attempts)); then
+      warn "Failed after $max_attempts attempts: $desc"
+      return 1
+    fi
+    info "Retry $attempt/$max_attempts for: $desc (sleeping ${sleep_sec}s)"
+    sleep "$sleep_sec"
+    ((attempt++))
+  done
+}
+
+# ── Wait for apt locks ───────────────────────────────────────────────
+# Brev VMs sometimes have unattended-upgrades running at boot.
+wait_for_apt_lock() {
+  local max_wait=120 elapsed=0
+  while fuser /var/lib/dpkg/lock-frontend >/dev/null 2>&1 \
+    || fuser /var/lib/apt/lists/lock >/dev/null 2>&1; do
+    if ((elapsed >= max_wait)); then
+      warn "apt lock not released after ${max_wait}s — proceeding anyway"
+      return 0
+    fi
+    if ((elapsed % 15 == 0)); then
+      info "Waiting for apt lock to be released... (${elapsed}s)"
+    fi
+    sleep 5
+    ((elapsed += 5))
+  done
+}
+
+# ══════════════════════════════════════════════════════════════════════
+# 1. System packages
+# ══════════════════════════════════════════════════════════════════════
+info "Installing system packages..."
+wait_for_apt_lock
+retry 3 10 "apt-get update" sudo apt-get update -qq
+retry 3 10 "apt-get install" sudo apt-get install -y -qq \
+  ca-certificates curl git jq tar >/dev/null 2>&1
+info "System packages installed"
+
+# ══════════════════════════════════════════════════════════════════════
+# 2. Docker
+# ══════════════════════════════════════════════════════════════════════
+if command -v docker >/dev/null 2>&1; then
+  info "Docker already installed"
+else
+  info "Installing Docker..."
+  wait_for_apt_lock
+  retry 3 10 "install docker" sudo apt-get install -y -qq docker.io >/dev/null 2>&1
+  info "Docker installed"
+fi
+sudo systemctl enable --now docker
+sudo usermod -aG docker "$TARGET_USER" 2>/dev/null || true
+# Make the socket world-accessible so SSH sessions (which don't pick up the
+# new docker group until re-login) can use Docker immediately.  This is a
+# short-lived CI VM — socket security is not a concern.
+sudo chmod 666 /var/run/docker.sock
+info "Docker enabled ($(docker --version 2>/dev/null | head -c 40))"
+
+# ══════════════════════════════════════════════════════════════════════
+# 3. Node.js 22
+# ══════════════════════════════════════════════════════════════════════
+node_major=""
+if command -v node >/dev/null 2>&1; then
+  node_major="$(node -p 'process.versions.node.split(".")[0]' 2>/dev/null || true)"
+fi
+
+if command -v npm >/dev/null 2>&1 && [[ -n "$node_major" ]] && ((node_major >= 22)); then
+  info "Node.js already installed: $(node --version)"
+else
+  info "Installing Node.js 22..."
+  curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash - >/dev/null 2>&1
+  wait_for_apt_lock
+  retry 3 10 "install nodejs" sudo apt-get install -y -qq nodejs >/dev/null 2>&1
+  info "Node.js $(node --version) installed"
+fi
+
+# ══════════════════════════════════════════════════════════════════════
+# 4. OpenShell CLI
+# ══════════════════════════════════════════════════════════════════════
+if command -v openshell >/dev/null 2>&1; then
+  info "OpenShell CLI already installed: $(openshell --version 2>&1 || echo unknown)"
+else
+  info "Installing OpenShell CLI ${OPENSHELL_VERSION}..."
+  ARCH="$(uname -m)"
+  case "$ARCH" in
+    x86_64 | amd64) ASSET="openshell-x86_64-unknown-linux-musl.tar.gz" ;;
+    aarch64 | arm64) ASSET="openshell-aarch64-unknown-linux-musl.tar.gz" ;;
+    *) fail "Unsupported architecture: $ARCH" ;;
+  esac
+  tmpdir="$(mktemp -d)"
+  retry 3 10 "download openshell" \
+    curl -fsSL -o "$tmpdir/$ASSET" \
+    "https://github.com/NVIDIA/OpenShell/releases/download/${OPENSHELL_VERSION}/${ASSET}"
+  tar xzf "$tmpdir/$ASSET" -C "$tmpdir"
+  sudo install -m 755 "$tmpdir/openshell" /usr/local/bin/openshell
+  rm -rf "$tmpdir"
+  info "OpenShell CLI installed: $(openshell --version 2>&1 || echo unknown)"
+fi
+
+# ══════════════════════════════════════════════════════════════════════
+# 5. Clone NemoClaw and install deps
+# ══════════════════════════════════════════════════════════════════════
+if [[ -d "$NEMOCLAW_CLONE_DIR/.git" ]]; then
+  info "NemoClaw repo exists at $NEMOCLAW_CLONE_DIR — refreshing"
+  git -C "$NEMOCLAW_CLONE_DIR" fetch origin "$NEMOCLAW_REF"
+  git -C "$NEMOCLAW_CLONE_DIR" checkout "$NEMOCLAW_REF"
+  git -C "$NEMOCLAW_CLONE_DIR" pull --ff-only origin "$NEMOCLAW_REF" || true
+else
+  info "Cloning NemoClaw (ref: $NEMOCLAW_REF)..."
+  git clone --branch "$NEMOCLAW_REF" --depth 1 \
+    "https://github.com/NVIDIA/NemoClaw.git" "$NEMOCLAW_CLONE_DIR"
+fi
+
+info "Installing npm dependencies..."
+cd "$NEMOCLAW_CLONE_DIR"
+npm install --ignore-scripts 2>&1 | tail -3
+info "Root deps installed"
+
+info "Building TypeScript plugin..."
+cd "$NEMOCLAW_CLONE_DIR/nemoclaw"
+npm install 2>&1 | tail -3
+npm run build 2>&1 | tail -3
+cd "$NEMOCLAW_CLONE_DIR"
+info "Plugin built"
+
+# ══════════════════════════════════════════════════════════════════════
+# 6. Pre-pull Docker images
+# ══════════════════════════════════════════════════════════════════════
+if [[ "${SKIP_DOCKER_PULL:-0}" == "1" ]]; then
+  info "Skipping Docker image pre-pulls (SKIP_DOCKER_PULL=1)"
+else
+  info "Pre-pulling Docker images (this saves 3-5 min per CI run)..."
+
+  # Use sg docker to ensure docker group is active without re-login
+  for image in "${DOCKER_IMAGES[@]}"; do
+    info "  Pulling $image..."
+    sg docker -c "docker pull $image" 2>&1 | tail -1 \
+      || warn "  Failed to pull $image (will be pulled at test time)"
+  done
+
+  # The openshell/cluster image tag should match the CLI version.
+  # Try the pinned version first, fall back to latest.
+  CLUSTER_TAG="${OPENSHELL_VERSION#v}" # v0.0.20 → 0.0.20
+  CLUSTER_IMAGE="ghcr.io/nvidia/openshell/cluster:${CLUSTER_TAG}"
+  info "  Pulling $CLUSTER_IMAGE..."
+  if ! sg docker -c "docker pull $CLUSTER_IMAGE" 2>&1 | tail -1; then
+    warn "  Could not pull $CLUSTER_IMAGE — trying :latest"
+    sg docker -c "docker pull ghcr.io/nvidia/openshell/cluster:latest" 2>&1 | tail -1 \
+      || warn "  Failed to pull openshell/cluster (will be pulled at test time)"
+  fi
+
+  info "Docker images pre-pulled"
+fi
+
+# ══════════════════════════════════════════════════════════════════════
+# 7. NVIDIA Container Toolkit
+# ══════════════════════════════════════════════════════════════════════
+if command -v nvidia-smi >/dev/null 2>&1; then
+  if ! dpkg -s nvidia-container-toolkit >/dev/null 2>&1; then
+    info "Installing NVIDIA Container Toolkit..."
+    curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey \
+      | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+    curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list \
+      | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' \
+      | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list >/dev/null
+    wait_for_apt_lock
+    retry 3 10 "apt-get update" sudo apt-get update -qq >/dev/null 2>&1
+    retry 3 10 "install nvidia-container-toolkit" sudo apt-get install -y -qq nvidia-container-toolkit >/dev/null 2>&1
+    sudo nvidia-ctk runtime configure --runtime=docker >/dev/null 2>&1
+    sudo systemctl restart docker
+    info "NVIDIA Container Toolkit installed"
+  else
+    info "NVIDIA Container Toolkit already installed"
+  fi
+
+  info "Validating GPU passthrough..."
+  if command -v nvidia-smi >/dev/null 2>&1; then
+    nvidia-smi || fail "GPU detected but not functioning"
+    docker run --rm --gpus all nvidia/cuda:12.2.0-base nvidia-smi \
+      || fail "Docker GPU passthrough failed"
+  else
+    fail "No GPU detected"
+  fi
+else
+  fail "nvidia-smi not found, assuming a CPU-only instance or failed passthrough"
+fi
+
+# ══════════════════════════════════════════════════════════════════════
+# 8. Ollama
+# ══════════════════════════════════════════════════════════════════════
+if ! command -v ollama >/dev/null 2>&1; then
+  info "Installing Ollama..."
+  curl -fsSL https://ollama.com/install.sh | sh >/dev/null 2>&1
+  info "Ollama installed"
+else
+  info "Ollama already installed"
+fi
+
+info "Pulling Ollama model $OLLAMA_MODEL..."
+if systemctl is-active --quiet ollama || pgrep -x ollama >/dev/null; then
+  ollama pull "$OLLAMA_MODEL" >/dev/null 2>&1 || warn "Failed to pull $OLLAMA_MODEL (might fallback to qwen2.5:0.5b)"
+  if ! ollama list | grep -q "${OLLAMA_MODEL}"; then
+    # fallback if qwen3:0.6b doesn't exist yet
+    if [ "$OLLAMA_MODEL" == "qwen3:0.6b" ]; then
+      ollama pull "qwen2.5:0.5b" >/dev/null 2>&1 || true
+    fi
+  fi
+else
+  # Start Ollama briefly to pull model if not running
+  ollama serve >/dev/null 2>&1 &
+  OLLAMA_PID=$!
+  info "Waiting for Ollama to start..."
+  poll_timeout=60
+  poll_elapsed=0
+  until curl -s http://localhost:11434 >/dev/null; do
+    if ((poll_elapsed >= poll_timeout)); then
+      fail "Ollama failed to start within ${poll_timeout}s"
+    fi
+    sleep 1
+    ((poll_elapsed++))
+  done
+  ollama pull "$OLLAMA_MODEL" >/dev/null 2>&1 || warn "Failed to pull $OLLAMA_MODEL (might fallback to qwen2.5:0.5b)"
+  if ! ollama list | grep -q "${OLLAMA_MODEL}"; then
+    if [ "$OLLAMA_MODEL" == "qwen3:0.6b" ]; then
+      ollama pull "qwen2.5:0.5b" >/dev/null 2>&1 || true
+    fi
+  fi
+  kill $OLLAMA_PID 2>/dev/null || true
+fi
+info "Ollama model pulling script complete."
+
+# ══════════════════════════════════════════════════════════════════════
+# 9. Readiness sentinel
+# ══════════════════════════════════════════════════════════════════════
+info "Running final validation..."
+
+docker ps >/dev/null || fail "Docker not working"
+node -v >/dev/null || fail "Node not working"
+
+if command -v ollama >/dev/null; then
+  ollama list >/dev/null || warn "Ollama not responding"
+fi
+
+info "Validation complete"
+
+sudo touch "$SENTINEL"
+echo "=== Ready ===" | sudo tee -a "$LAUNCH_LOG" >/dev/null
+
+info "════════════════════════════════════════════════════"
+info "  CI-Ready GPU launchable setup complete"
+info "  NemoClaw:   $NEMOCLAW_CLONE_DIR (ref: $NEMOCLAW_REF)"
+info "  OpenShell:  $(openshell --version 2>&1 || echo unknown)"
+info "  Node.js:    $(node --version)"
+info "  Docker:     $(docker --version 2>/dev/null | head -c 40)"
+info "  Ollama:     $(ollama --version 2>/dev/null || echo unknown)"
+info "  Sentinel:   $SENTINEL"
+info "════════════════════════════════════════════════════"
diff --git a/test/e2e/brev-e2e.test.js b/test/e2e/brev-e2e.test.js
index ce67e8943..a18ade991 100644
--- a/test/e2e/brev-e2e.test.js
+++ b/test/e2e/brev-e2e.test.js
@@ -17,7 +17,7 @@
  *   INSTANCE_NAME    — Brev instance name (e.g. pr-156-test)
  *
  * Optional env vars:
- *   TEST_SUITE             — which test to run: full (default), credential-sanitization, telegram-injection, all
+ *   TEST_SUITE             — which test to run: full (default), credential-sanitization, telegram-injection, all, gpu
  *   USE_LAUNCHABLE         — "1" (default) to use CI launchable, "0" for bare brev create + brev-setup.sh
  *   LAUNCHABLE_SETUP_SCRIPT — URL to setup script for launchable path (default: brev-launchable-ci-cpu.sh on main)
  *   BREV_MIN_VCPU          — Minimum vCPUs for CPU instance (default: 4)