diff --git a/.claude/machines/guru-kali.md b/.claude/machines/guru-kali.md
index 4ed0359..fe56a56 100644
--- a/.claude/machines/guru-kali.md
+++ b/.claude/machines/guru-kali.md
@@ -1,7 +1,7 @@
 # Machine: GURU-KALI
 
 **Hostname:** GURU-KALI
-**Last Updated:** 2026-05-24
+**Last Updated:** 2026-05-25
 
 ---
 
@@ -32,7 +32,7 @@
 | mysql/mariadb client | 11.8.6 |
 | nmap | 7.99 (Kali security tooling) |
 | GuruRMM build dev libs | libgtk-3-dev, libayatana-appindicator3-dev, libxdo-dev, libssl-dev, pkg-config (for agent + tray builds) — added 2026-05-24 |
-| NVIDIA driver | nouveau (open-source) — NO proprietary driver / CUDA |
+| NVIDIA driver | nvidia-open 595.71.05 (open kernel modules + CUDA, DKMS) — built/signed for kernel 6.19 via the NVIDIA CUDA debian13 repo; nouveau blacklisted; **ACTIVE AFTER REBOOT**. Added 2026-05-25. |
 | jq | 1.8.1 (added 2026-05-24, needed by hooks) |
 | gh / docker / age / op / grepai / ollama | NOT installed |
 
@@ -42,8 +42,8 @@
 
 None — Ollama not installed. If installed, `qwen3:8b` (5.2 GB) fits the 8 GB VRAM
 fully (mirrors DESKTOP-0O8A1RL prose model); qwen3.6 / codestral / qwen3:14b would
-split to CPU. GPU acceleration requires the proprietary NVIDIA driver first (currently
-nouveau). See `.claude/OLLAMA.md`.
+split to CPU. GPU acceleration is available via nvidia-open 595.71.05 (CUDA) once the
+pending reboot activates the driver. See `.claude/OLLAMA.md`.
 
 ---
 
@@ -97,7 +97,7 @@ Verified 2026-05-24: coord API `172.16.3.30:8001` -> HTTP 200, remote Ollama
 - [ ] Ollama LOCAL (not installed — would add offline Tier 0)
 - [ ] GrepAI semantic search (not installed)
 - [ ] 1Password CLI (op not installed)
-- [ ] NVIDIA CUDA compute (nouveau driver — no CUDA)
+- [~] NVIDIA CUDA compute — nvidia-open 595.71.05 installed (DKMS built on kernel 6.19); ACTIVE after reboot
 - [ ] Docker
 
 ---
@@ -105,8 +105,8 @@ Verified 2026-05-24: coord API `172.16.3.30:8001` -> HTTP 200, remote Ollama
 ## Notes
 
 - **Strongest raw hardware in the fleet** for AI inference (i9-14900HX, 31 GB RAM,
-  RTX 4070 8 GB) — but GPU compute is blocked on the nouveau -> proprietary-NVIDIA
-  driver swap (needs a package install + reboot on Kali rolling).
+  RTX 4070 8 GB). NVIDIA driver now installed (see 2026-05-25 note) — GPU/CUDA compute
+  available after the pending reboot.
 - **Field/mobile laptop.** On wifi off the company LAN, but Tailscale (added
   2026-05-24) bridges to internal services, so coord API/DB and remote Ollama work.
   A local Ollama would still add value for *offline* use (away from any network).
@@ -124,3 +124,19 @@ Verified 2026-05-24: coord API `172.16.3.30:8001` -> HTTP 200, remote Ollama
   Screen may still blank/screensave on idle but does NOT prompt for a password. Do NOT
   re-enable. Persisted in `~/.config/xfce4/xfconf/xfce-perchannel-xml/xfce4-screensaver.xml`
   (machine-local, not in the repo).
+- **2026-05-24: idle-suspend (sleep) DISABLED on AC *and* battery (user request).**
+  xfce4-power-manager `inactivity-on-ac=0`, `inactivity-on-battery=0` (+ sleep-mode actions 0).
+  Why both: it was sleeping while plugged in because the ACPI AC-adapter state goes **stale** on
+  this Legion + kernel 6.19 (the EC doesn't reliably send AC plug/unplug events, esp. across S3
+  resume), so the OS intermittently thought it was on battery and applied the 10-min battery
+  suspend. `udevadm trigger --action=change /sys/class/power_supply/ADP0` forces a re-read
+  (online flips back to 1). Disabling idle-suspend on both states makes the flaky detection
+  harmless. Do NOT re-enable. Persisted in `xfce4-power-manager.xml` (machine-local). Proper
+  upstream fix = Lenovo BIOS/EC update.
+- **2026-05-25: NVIDIA driver installed — nvidia-open 595.71.05** (open kernel modules + CUDA),
+  via the NVIDIA CUDA `debian13` repo (cuda-keyring) since Kali only packages 550 (too old for
+  kernel 6.19). DKMS module built + signed for `6.19.14+kali-amd64`; nouveau blacklisted
+  (`/etc/modprobe.d/nvidia.conf`) + initramfs rebuilt. **Requires one reboot to activate** (swap
+  the dGPU off nouveau). Hybrid graphics: Intel i915 stays the display, NVIDIA dGPU for
+  CUDA/offload. Secure Boot off. Verify post-reboot: `nvidia-smi` (expect RTX 4070 Mobile, driver
+  595.71.05). Unblocks the local-Ollama-on-GPU option.