llm, add Swap Stopper as a service based on custom shell script.

files/scripts/llm_swap_stopper.sh

@@ -0,0 +1,84 @@
+#!/bin/sh -e
+# 2026-10-10 Hyperling
+# Stop a system from going into swap due to LLMs.
+
+### Tuning ###
+
+typeset -i sleep=10 ram_pct_restart=98
+
+### Functions ###
+
+restart_service() {
+	if [ -z $1 ]; then
+		echo "ERROR: Service name must be passed to restart_service()."
+		return 1
+	fi
+	sv restart $1 || echo "$1 failed to restart, ignoring."
+}
+
+# While we could loop over `ollama ps` and stop them, this ensures that the UIs
+#   do not sit around waiting for responses from the dead LLMs. It bring them up
+#   in a fresh state so that work can continue from a more clean slate.
+restart_llms() {
+	restart_service ollama
+	restart_service zeroclaw
+	restart_service open-webui
+}
+
+### Main ###
+
+typeset -i {total,used}_{ram,swap}=0 \
+	used_{ram,swap}_pct=0 \
+	last_{ram,swap}_pct=100 \
+	count=0
+
+while [[ true ]]; do
+
+	count=$(expr $count + 1 )
+	echo "$(date +'%Y-%m-%d @ %H:%M:%S') - Loop $count"
+
+	# RAM #
+
+	total_ram=$(free | awk '/^Mem:/ {print $2}')
+	used_ram=$(free | awk '/^Mem:/ {print $3}')
+	used_ram_pct=$( \
+		awk -v used="$used_ram" -v total="$total_ram" \
+		'BEGIN {printf "%d",(used/total)*100}' \
+	)
+
+	echo "Current RAM usage is $used_ram_pct%, was $last_ram_pct% last loop."
+
+	# Check if RAM is over 95%
+	if [ $used_ram_pct -gt $ram_pct_restart ]; then
+		echo "WARNING: RAM is over ${ram_pct_restart}% used. Restarting LLMs."
+		restart_llms
+		continue
+	fi
+
+	# Swap #
+
+	total_swap=$(free | awk '/^Swap:/ {print $2}')
+	used_swap=$(free | awk '/^Swap:/ {print $3}')
+	used_swap_pct=$( \
+		awk -v used="$used_swap" -v total="$total_swap" \
+		'BEGIN {printf "%d",(used/total)*100}' \
+	)
+
+	echo "Current swap usage is $used_swap_pct%, was $last_swap_pct% last loop."
+
+	# Check if swap is growing
+	if [ $used_swap_pct -gt $last_swap_pct ]; then
+		echo "WARNING: Swap is growing. Restarting LLMs."
+		restart_llms
+		continue
+	fi
+
+	# Finish #
+
+	echo -e "Check finished, sleeping for ${sleep}...\n"
+
+	last_ram_pct=$used_ram_pct
+	last_swap_pct=$used_swap_pct
+
+	sleep $sleep
+done

local.yml

@@ -2,7 +2,7 @@
 # Harmonize systems rather than doing everything manually. :)
 
 # Setup of any device connected to this repo.
-- name: Main | Default Setup
+- name: Local | Main
   hosts: localhost
   connection: local
   become: true
@@ -10,13 +10,14 @@
   tasks:
 
     ####### Header #######
+
     - name: Main | Header
       ansible.builtin.include_tasks: tasks/header.yml
 
-
     ####### Workstations #######
+
     # Additional setup for systems with GUI.
-    - name: Main | Workstation Setup
+    - name: Main | Workstation
       block:
 
         - ansible.builtin.include_tasks: facts/workstation/package.yml
@@ -30,6 +31,7 @@
         # Window Managers without true Desktop Environments #
         - ansible.builtin.include_tasks: tasks/workstation/shared/software/dwm.yml
           when: internet
+
         - ansible.builtin.include_tasks: tasks/workstation/shared/software/i3.yml
           when: internet
 
@@ -43,7 +45,9 @@
 
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/gnome.yml
               when: ansible_service_mgr == "systemd"
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/cinnamon.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/lxqt.yml
 
           when: ansible_system == "Linux"
@@ -52,16 +56,23 @@
 
         ## Software Tasks ##
 
-        - name: Main | Workstation Setup | Linux
+        - name: Main | Workstation | Linux
           block:
 
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/packages.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/tlp.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/lutris.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/vscodium.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/vpn.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/nodejs.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/hugo.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/brave.yml
               when: ansible_pkg_mgr in ("apt", "dnf", "zypper") and not mobile
 
@@ -77,11 +88,14 @@
             and internet
 
         ## Configuration Tasks ##
-        - name: Main | Workstation Setup | Configuration
+
+        - name: Main | Workstation | Configuration
           block:
 
             - ansible.builtin.include_tasks: tasks/workstation/shared/settings/gnome.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/shared/settings/cinnamon.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/shared/settings/xdg_autostart.yml
 
             - ansible.builtin.include_tasks: tasks/workstation/linux/cron/ansible.yml
@@ -98,17 +112,23 @@
           when: vnc
 
         - ansible.builtin.include_tasks: tasks/workstation/shared/settings/services.yml
+
         - ansible.builtin.include_tasks: tasks/workstation/shared/settings/templates.yml
 
         ## Final Tasks (SLOW) ##
-        - name: Main | Workstation Setup | Linux [Slow]
+
+        - name: Main | Workstation | Linux [Slow]
           block:
 
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/flatpaks.yml
               when: flatpak_distro
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/java.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/flutter.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/android.yml
+
             - ansible.builtin.include_tasks: tasks/workstation/linux/software/llm.yml
 
           when: ansible_system == "Linux"
@@ -116,12 +136,11 @@
 
       when: workstation
 
-
 # No longer mining, this is now considered deprecated.
 #    ####### Miners #######
 #
 #    # Additional setup for miners.
-#    - name: Main | Miner Setup
+#    - name: Main | Miner
 #      block:
 #
 #      ##### Setup #####
@@ -157,32 +176,36 @@
 #
 #      when: miner
 
-
     ####### Servers #######
+
     # Easy to deploy server configurations.
-    - name: Main | Server Setup
+    - name: Main | Server
       block:
 
-        - name: Main | Server Setup | Services
+        - name: Main | Server | Services
           ansible.builtin.include_tasks: tasks/server/software/services.yml
 
         - block:
 
-            - name: Main | Server Setup | Certbot
+            - name: Main | Server | Certbot
               ansible.builtin.include_tasks: tasks/server/software/certbot.yml
-            - name: Main | Server Setup | Certbot Cron
+
+            - name: Main | Server | Certbot Cron
               ansible.builtin.include_tasks: tasks/server/cron/certbot.yml
               when: certbot
 
         - block:
 
-            - name: Main | Server Setup | OnlyOffice
+            - name: Main | Server | OnlyOffice
               ansible.builtin.include_tasks: tasks/server/software/onlyoffice.yml
               when: onlyoffice
+
             - ansible.builtin.include_tasks: tasks/server/software/influxdb1.yml
               when: influxdb1
+
             - ansible.builtin.include_tasks: tasks/server/software/influxdb2.yml
               when: influxdb2
+
             - ansible.builtin.include_tasks: tasks/server/software/grafana.yml
               when: grafana
 
@@ -193,8 +216,8 @@
 
       when: server
 
-
     ####### Reporting #######
+
     # Provide information for analysis.
     - name: Main | Reporting
       block:
@@ -203,10 +226,12 @@
           when: internet
 
         - ansible.builtin.include_tasks: tasks/general/tests/services.yml
+
         - ansible.builtin.include_tasks: tasks/general/tests/lynis.yml
+
         - ansible.builtin.include_tasks: tasks/general/tests/nmap.yml
 
-
     ####### Footer #######
+
     - name: Main | Footer
       ansible.builtin.include_tasks: tasks/footer.yml

tasks/footer.yml

@@ -1,4 +1,5 @@
 ---
+# All the tasks which need run to successfully end the playbook.
 
 ####### Footer #######
 - name: Footer
@@ -9,4 +10,5 @@
       vars:
         notify_message: "Provisioning Finished"
 
-    - ansible.builtin.include_tasks: tasks/general/tests/lock_remove.yml
+    - name: Footer | Lock | Remove
+      ansible.builtin.include_tasks: tasks/general/tests/lock_remove.yml

tasks/general/acct_mgmt/users.yml

@@ -1146,6 +1146,7 @@
           sleep='echo -e "`date` - Sleeping for $sleep_time... \n"; sleep $sleep_time'
         fi
         count=0
+        echo "Command='$cmd'"
         while true; do
           (( count++ ))
           echo "`date` - Loop $count"

tasks/header.yml

@@ -1,40 +1,46 @@
 ---
 
 ####### Lock #######
+
 # Ensure a process is not already running.
 - name: Main | Lock
   block:
 
     - name: Main | Lock | Check
       ansible.builtin.include_tasks: tasks/general/tests/lock_check.yml
+
     - name: Main | Lock | Create
       ansible.builtin.include_tasks: tasks/general/tests/lock_create.yml
 
-
 ####### Gather #######
+
 # Load facts for all systems.
 - name: Main | Gather
   block:
 
     - name: Main | Gather | Facts
       ansible.builtin.include_tasks: facts/general/gather.yml
+
     - name: Main | Gather | Account Management
       ansible.builtin.include_tasks: tasks/general/acct_mgmt/provision_config.yml
 
     - name: Main | Gather | System
       ansible.builtin.include_tasks: facts/general/system.yml
+
     - name: Main | Gather | Package
       ansible.builtin.include_tasks: facts/general/package.yml
+
     - name: Main | Gather | Service
       ansible.builtin.include_tasks: facts/general/service.yml
+
     - name: Main | Gather | User
       ansible.builtin.include_tasks: facts/general/user.yml
 
     - name: Main | Gather | Internet Check
       ansible.builtin.include_tasks: tasks/general/tests/internet_check.yml
 
-
 ####### Start #######
+
 - name: Main | Start
   block:
 
@@ -46,25 +52,31 @@
   when: workstation
 
 ####### NixOS #######
+
 # Install the .nix files and do a little data setup.
-- name: Main | NixOS Setup
+- name: Main | NixOS
   block:
 
-    - name: Main | NixOS Setup | NixOS
+    - name: Main | NixOS | Config
       ansible.builtin.include_tasks: tasks/general/software/nixos.yml
 
     # TBD if we are going to use this many, then just have IF statements
     #  and use blocks within the General area for NixOS vs not-NixOS.
-    - name: Main | NixOS Setup | Swap
+    - name: Main | NixOS | Swap
       ansible.builtin.include_tasks: tasks/general/software/swap.yml
 
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/groups.yml
+
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/users.yml
+
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/sudo.yml
+
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/doas.yml
 
     - ansible.builtin.include_tasks: tasks/general/scripts/root.yml
+
     - ansible.builtin.include_tasks: tasks/general/scripts/user.yml
+
     - ansible.builtin.include_tasks: tasks/general/scripts/install.yml
 
     - ansible.builtin.include_tasks: tasks/general/cron/ansible.yml
@@ -78,34 +90,36 @@
     # Try including these. Really need to get NixOS working with the rest!
     #- ansible.builtin.include_tasks: tasks/general/software/telegraf.yml # "Could not find a module for unknown.""
     - ansible.builtin.include_tasks: tasks/general/tests/services.yml
+
     - ansible.builtin.include_tasks: tasks/general/tests/lynis.yml
+
     - ansible.builtin.include_tasks: tasks/general/tests/nmap.yml
 
-    - name: NixOS | Complete
+    - name: Main | NixOS | Complete
       ansible.builtin.meta: end_play
 
   when: ansible_distribution == "NixOS"
 
-
 ####### General #######
+
 # Basic setup for all enabled systems.
-- name: Main | General Setup
+- name: Main | General
   block:
 
-    - name: Main | General Setup | Swap
+    - name: Main | General | Swap
       ansible.builtin.include_tasks: tasks/general/software/swap.yml
 
-    - name: Main | General Setup | runit
+    - name: Main | General | runit
       ansible.builtin.include_tasks: tasks/general/software/runit.yml
 
-    - name: Main | General Setup | Packages
+    - name: Main | General | Packages
       ansible.builtin.include_tasks: tasks/general/software/packages.yml
       when: internet
 
-    - name: Main | General Setup | Services
+    - name: Main | General | Services
       ansible.builtin.include_tasks: tasks/general/software/services.yml
 
-    - name: Main | General Setup | Sendmail
+    - name: Main | General | Sendmail
       ansible.builtin.include_tasks: tasks/general/software/sendmail.yml
       when: ansible_system == "FreeBSD"
         and internet
@@ -119,8 +133,11 @@
     - ansible.builtin.include_tasks: tasks/general/settings/keyboard.yml
 
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/groups.yml
+
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/users.yml
+
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/sudo.yml
+
     - ansible.builtin.include_tasks: tasks/general/acct_mgmt/doas.yml
       when: internet
 
@@ -128,7 +145,9 @@
       when: internet
 
     - ansible.builtin.include_tasks: tasks/general/scripts/root.yml
+
     - ansible.builtin.include_tasks: tasks/general/scripts/user.yml
+
     - ansible.builtin.include_tasks: tasks/general/scripts/install.yml
 
     - ansible.builtin.include_tasks: tasks/general/cron/ansible.yml

tasks/workstation/linux/software/cinnamon.yml

@@ -21,4 +21,3 @@
       - "{{ cinnamon_package }}"
       - "{{ lightdm }}"
     state: present
-    update_cache: true

tasks/workstation/linux/software/flatpaks.yml

@@ -19,7 +19,6 @@
 - name: Workstation | Linux | Flatpak Distro | Flatpak | Install
   ansible.builtin.package:
     name: flatpak
-    update_cache: true
 
 - name: Workstation | Linux | Flatpak Distro | Flatpak | Add Flathub for System
   flatpak_remote:

tasks/workstation/linux/software/llm.yml

@@ -26,6 +26,8 @@
 - name: Workstation | Linux | Software | LLMs
   block:
 
+    # Facts - Start #
+
     - name: Workstation | Linux | Software | LLMs | Facts | Provisioned
       ansible.builtin.debug:
         var: "{{ item }}"
@@ -47,10 +49,12 @@
         qwen_code: qwen-code
         dummy_service: 'cronie'
         dummy_website: 'https://hyperling.com'
-        # Good to keep low to prevent swap usage.
+        # Good to keep low to prevent swap usage. Going to have a minimum of 3
-        llm_concurrency: "{{ (ansible_memtotal_mb / 64000) | round(0, method='ceil') | int }}"
+        # though, one for Cline, two for Continue (Agent + Autocomplete).
+        llm_concurrency: "{{ [(ansible_memtotal_mb / 32000) | round(0, method='ceil'), 3] | max | int }}"
         # Allow having multiple chats open without them just waiting each other.
-        llm_parallel: "{{ (ansible_processor_vcpus / 8) | round(0, method='ceil') | int }}"
+        # Also setting this to a minimum of 3 for using Codium with extensions.
+        llm_parallel: "{{ [(ansible_processor_vcpus / 2) | round(0, method='ceil'), 3] | max | int }}"
         # Most LLMs take a coefficient of about 1.2 to 1.5 of their file size in
         #    RAM, which is usually less than guessing a usage of weight*1000.
         # Example:
@@ -58,8 +62,8 @@
         # Conversely:
         #   - qwen3.5:4b would be 3.3GB * 1.3 = 4290, rather than 4000.
         llm_size_coef: 1.3
-        # Take the GB of RAM and use it as the MB of context.
+        # Use the MB of RAM as the context, with a maximum to prevent swapping.
-        llm_context: "{{ ansible_memtotal_mb | int }}"
+        llm_context: "{{ [(ansible_memtotal_mb | round(-3, method='ceil')), 32000] | min | int }}"
 
     - name: Workstation | Linux | Software | LLMs | Facts 02
       ansible.builtin.set_fact:
@@ -67,6 +71,8 @@
         ollama_runit_log_dest: "{{ global_log }}/{{ ollama }}"
         zeroclaw_runit_dir: "{{ runit_script_dir }}/{{ zeroclaw }}"
         zeroclaw_runit_log_dest: "{{ global_log }}/{{ zeroclaw }}"
+        # Amount of output tokens allowed per response. Half of context is ok.
+        llm_tokens: "{{ (llm_context / 4) | int }}"
 
     - name: Workstation | Linux | Software | LLMs | Facts 03
       ansible.builtin.set_fact:
@@ -86,33 +92,33 @@
       ansible.builtin.set_fact:
         llm_installs: []
         llm_removals:
-          - { "model": "qwen3:4b", "provider": "qwen_30_0040", "size_mb": 4000 }
+          - { "purpose": "chat", "model": "qwen3:4b", "provider": "qwen_30_0040", "size_mb": 4000, "context": 123456 }
-          - { "model": "qwen3.6:27b", "provider": "qwen_36_0270", "size_mb": 27000 }
+          - { "purpose": "chat", "model": "qwen3.6:27b", "provider": "qwen_36_0270", "size_mb": 27000, "context": 123456 }
-          - { "model": "qwen3.5:27b", "provider": "qwen_35_0270", "size_mb": "{{ llm_size_coef * 17000 }}" }
+          - { "purpose": "chat", "model": "qwen3.5:27b", "provider": "qwen_35_0270", "size_mb": "{{ llm_size_coef * 17000 }}", "context": 123456 }
         llm_choices:
             ### QWEN >3.5 - Need to try!
-          - { "model": "qwen3.8:27b", "provider": "qwen_38_0270", "size_mb": 27000 }
+          - { "purpose": ["chat", code], "model": "qwen3.8:27b", "provider": "qwen_38_0270", "size_mb": 27000, "context": 123456 }
-          - { "model": "qwen3.6:35b", "provider": "qwen_36_0350", "size_mb": 35000 }
+          - { "purpose": ["chat", code], "model": "qwen3.6:35b", "provider": "qwen_36_0350", "size_mb": 35000, "context": 123456 }
             ### QWEN 3.5 - Good for generic usage and a little coding.
-          - { "model": "qwen3.5:0.8b", "provider": "qwen_35_0008", "size_mb": "{{ llm_size_coef *  1200 }}" }
+          - { "purpose": "chat", "model": "qwen3.5:0.8b", "provider": "qwen_35_0008", "size_mb": "{{ llm_size_coef *  1200 }}", "context": 256000 }
-          - { "model": "qwen3.5:2b",   "provider": "qwen_35_0020", "size_mb": "{{ llm_size_coef *  2700 }}" }
+          - { "purpose": "chat", "model": "qwen3.5:2b",   "provider": "qwen_35_0020", "size_mb": "{{ llm_size_coef *  2700 }}", "context": 256000 }
-          - { "model": "qwen3.5:4b",   "provider": "qwen_35_0040", "size_mb": "{{ llm_size_coef *  3300 }}" }
+          - { "purpose": "chat", "model": "qwen3.5:4b",   "provider": "qwen_35_0040", "size_mb": "{{ llm_size_coef *  3300 }}", "context": 256000 }
-          - { "model": "qwen3.5:9b",   "provider": "qwen_35_0090", "size_mb": "{{ llm_size_coef *  6600 }}" }
+          - { "purpose": "chat", "model": "qwen3.5:9b",   "provider": "qwen_35_0090", "size_mb": "{{ llm_size_coef *  6600 }}", "context": 256000 }
               # skip 27b and use 35b
-          - { "model": "qwen3.5:35b",  "provider": "qwen_35_0350", "size_mb": "{{ llm_size_coef * 22000 }}" }
+          - { "purpose": ["chat", code], "model": "qwen3.5:35b",  "provider": "qwen_35_0350", "size_mb": "{{ llm_size_coef * 22000 }}", "context": 256000 }
-          - { "model": "qwen3.5:122b", "provider": "qwen_35_1220", "size_mb": "{{ llm_size_coef * 81000 }}" }
+          - { "purpose": ["chat", code], "model": "qwen3.5:122b", "provider": "qwen_35_1220", "size_mb": "{{ llm_size_coef * 81000 }}", "context": 256000 }
             ### QWEN 3.0 Coder - Supposed to be better than 2.5 for machines that can handle it.
-          - { "model": "qwen3-coder:30b",  "provider": "qwen_30_coder_0300", "size_mb": "{{ llm_size_coef *  19000 }}" }
+          - { "purpose": "code", "model": "qwen3-coder:30b",  "provider": "qwen_30_coder_0300", "size_mb": "{{ llm_size_coef *  19000 }}", "context": 123456 }
-          - { "model": "qwen3-coder:420b", "provider": "qwen_30_coder_4200", "size_mb": "{{ llm_size_coef * 290000 }}" }
+          - { "purpose": "code", "model": "qwen3-coder:420b", "provider": "qwen_30_coder_4200", "size_mb": "{{ llm_size_coef * 290000 }}", "context": 123456 }
             ### QWEN 2.5 Coder - Ol'reliable!
-          - { "model": "qwen2.5-coder:0.5b", "provider": "qwen_25_coder_0005", "size_mb": "{{ llm_size_coef *   398 }}" }
+          - { "purpose": "code", "model": "qwen2.5-coder:0.5b", "provider": "qwen_25_coder_0005", "size_mb": "{{ llm_size_coef *   398 }}", "context": 32000 }
-          - { "model": "qwen2.5-coder:1.5b", "provider": "qwen_25_coder_0015", "size_mb": "{{ llm_size_coef *   986 }}" }
+          - { "purpose": "code", "model": "qwen2.5-coder:1.5b", "provider": "qwen_25_coder_0015", "size_mb": "{{ llm_size_coef *   986 }}", "context": 32000 }
-          - { "model": "qwen2.5-coder:3b",   "provider": "qwen_25_coder_0030", "size_mb": "{{ llm_size_coef *  1900 }}" }
+          - { "purpose": "code", "model": "qwen2.5-coder:3b",   "provider": "qwen_25_coder_0030", "size_mb": "{{ llm_size_coef *  1900 }}", "context": 32000 }
-          - { "model": "qwen2.5-coder:7b",   "provider": "qwen_25_coder_0070", "size_mb": "{{ llm_size_coef *  4700 }}" }
+          - { "purpose": "code", "model": "qwen2.5-coder:7b",   "provider": "qwen_25_coder_0070", "size_mb": "{{ llm_size_coef *  4700 }}", "context": 32000 }
-          - { "model": "qwen2.5-coder:14b",  "provider": "qwen_25_coder_0140", "size_mb": "{{ llm_size_coef *  9000 }}" }
+          - { "purpose": "code", "model": "qwen2.5-coder:14b",  "provider": "qwen_25_coder_0140", "size_mb": "{{ llm_size_coef *  9000 }}", "context": 32000 }
-          - { "model": "qwen2.5-coder:32b",  "provider": "qwen_25_coder_0320", "size_mb": "{{ llm_size_coef * 20000 }}" }
+          - { "purpose": "code", "model": "qwen2.5-coder:32b",  "provider": "qwen_25_coder_0320", "size_mb": "{{ llm_size_coef * 20000 }}", "context": 32000 }
             ### Mistral, Ministral, and Devstral - By a French company?
-          - { "model": "devstral-small-2:24b", "provider": "devstral_20_0240_small", "size_mb": 24000 }
+          - { "purpose": "code", "model": "devstral-small-2:24b", "provider": "devstral_20_0240_small", "size_mb": 24000, "context": 123456 }
           #- mistral:7b
           #- ministral-3:3b
           #- ministral-3:8b
@@ -141,6 +147,9 @@
         - ansible_memtotal_mb
         - llm_concurrency
         - llm_parallel
+        - llm_size_coef
+        - llm_context
+        - llm_tokens
         - llm_installs
         - llm_removals
         - llm_choices
@@ -159,6 +168,8 @@
         - llm_installs
         - llm_removals
 
+    # Facts - End #
+
     - name: Workstation | Linux | Software | LLMs | Packages | Install
       ansible.builtin.package:
         name:
@@ -171,6 +182,8 @@
     - name: Workstation | Linux | Software | LLMs | runit
       block:
 
+        # Ollama - Start #
+
         - name: Workstation | Linux | Software | LLMs | runit | ollama | Dir | Run
           ansible.builtin.file:
             name: "{{ ollama_runit_dir }}"
@@ -208,7 +221,7 @@
               export OLLAMA_REQUEST_TIMEOUT=3600 # default 300 (unconfirmed, not in the list of --help vars)
 
               ### Resource Usage ###
-              export OLLAMA_KEEP_ALIVE=20m # default is 5m
+              export OLLAMA_KEEP_ALIVE=2h # default is 5m
               export OLLAMA_MAX_LOADED_MODELS={{ llm_concurrency }} # default is 3 if using CPUs, or 3 * NumGPUs
               export OLLAMA_NUM_PARALLEL={{ llm_parallel }} # default is 1 and applies per-model (unconfirmed)
               export OLLAMA_CONTEXT_LENGTH={{ llm_context }} # default is 4096, https://docs.ollama.com/faq#how-can-i-specify-the-context-window-size
@@ -245,6 +258,10 @@
             owner: "{{ user }}"
             mode: "0755"
 
+        # Ollama - End #
+
+        # Zeroclaw - Start #
+
         - name: Workstation | Linux | Software | LLMs | runit | zeroclaw | Dir | Run
           ansible.builtin.file:
             name: "{{ zeroclaw_runit_dir }}"
@@ -295,6 +312,9 @@
               dateformat %Y-%m-%d_%H-%M-%S # each run gets its own file.
             mode: "0755"
 
+        # Zeroclaw - End #
+
+        # Open WebUI - Start #
 
         - name: Workstation | Linux | Software | LLMs | runit | open-webui | Packages
           ansible.builtin.package:
@@ -325,7 +345,7 @@
             owner: "{{ user }}"
             mode: "0755"
 
-        - name: Workstation | Linux | Software | LLMs | runit | open-webui | Script | File
+        - name: Workstation | Linux | Software | LLMs | runit | open-webui | Script | Run
           ansible.builtin.copy:
             dest: /etc/sv/open-webui/run
             content: |
@@ -336,14 +356,14 @@
               while [ ! -S /var/run/docker.sock ]; do sleep 1; done
 
               # Ensure the old container is gone.
-              echo "Stopping..."
+              echo "RUN: Stopping..."
               docker stop open-webui &&
-                echo "Container stopped" ||
+                echo "RUN: Container stopped" ||
-                echo "Container already stopped."
+                echo "RUN: Container already stopped."
-              echo "Deleting..."
+              echo "RUN: Deleting..."
               docker rm -f open-webui &&
-                echo "Container deleted" ||
+                echo "RUN: Container deleted" ||
-                echo "Container already deleted."
+                echo "RUN: Container already deleted."
 
               # Start the open-webui program.
               set -e
@@ -358,6 +378,33 @@
                 ghcr.io/open-webui/open-webui:main
             mode: "0755"
 
+        - name: Workstation | Linux | Software | LLMs | runit | open-webui | Script | Finish
+          ansible.builtin.copy:
+            dest: /etc/sv/open-webui/finish
+            content: |
+              #!/bin/sh
+              exec 2>&1
+              # $1 is the exit code, $2 is the exit status byte (signal number if killed)
+              echo "FINISH: Service returned with exit code: $1 and status: $2"
+
+              echo "FINISH: Stopping..."
+              docker stop open-webui &&
+                echo "FINISH: Container stopped" ||
+                echo "FINISH: Container already stopped."
+              echo "FINISH: Deleting..."
+              docker rm -f open-webui &&
+                echo "FINISH: Container deleted" ||
+                echo "FINISH: Container already deleted."
+
+              sleep=30
+              if [[ "$1" != 0 ]]; then
+                echo "FINISH: Sleeping for $sleep seconds due to error..."
+                sleep $sleep
+              fi
+
+              echo "FINISH: Done!"
+            mode: "0755"
+
         - name: Workstation | Linux | Software | LLMs | runit | open-webui | Script | Log
           ansible.builtin.copy:
             dest: /etc/sv/open-webui/log/run
@@ -375,8 +422,72 @@
               dateformat %Y-%m-%d_%H-%M-%S # each run gets its own file.
             mode: "0755"
 
+        # Open WebUI - End #
+
+        # Swap Stopper - Start #
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Script | Executable
+          ansible.builtin.copy:
+            dest: "{{ global_bin }}/llm_swap_stopper.sh"
+            src: files/scripts/llm_swap_stopper.sh
+            mode: "0755"
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Dir | /etc/sv
+          ansible.builtin.file:
+            name: "/etc/sv/swap-stopper/log"
+            state: directory
+            owner: root
+            mode: "0755"
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Dir | /var/log
+          ansible.builtin.file:
+            name: "/var/log/swap-stopper"
+            state: directory
+            owner: "{{ user }}"
+            mode: "0755"
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Script | Run
+          ansible.builtin.copy:
+            dest: /etc/sv/swap-stopper/run
+            content: |
+              #!/bin/sh -e
+              exec 2>&1
+              exec {{ global_bin }}/llm_swap_stopper.sh
+            mode: "0755"
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Script | Finish
+          ansible.builtin.copy:
+            dest: /etc/sv/swap-stopper/finish
+            content: |
+              #!/bin/sh -e
+              exec 2>&1
+              # $1 is the exit code, $2 is the exit status byte (signal number if killed)
+              echo "FINISH: Service returned with exit code: $1 and status: $2"
+            mode: "0755"
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Script | Log
+          ansible.builtin.copy:
+            dest: /etc/sv/swap-stopper/log/run
+            content: |
+              #!/bin/sh -e
+              exec svlogd -tt /var/log/swap-stopper
+            mode: "0755"
+
+        - name: Workstation | Linux | Software | LLMs | runit | swap-stopper | Script | Log Config
+          ansible.builtin.copy:
+            dest: /var/log/swap-stopper/config
+            content: |
+              s1000000                     # rotate at 1 MB
+              n5                           # keep 5 old files
+              dateformat %Y-%m-%d_%H-%M-%S # each run gets its own file.
+            mode: "0755"
+
+        # Swap Saver - End #
+
       when: ansible_service_mgr == "runit"
 
+    # Services - Start #
+
     - name: Workstation | Linux | Software | LLMs | Services | Enable
       ansible.builtin.service:
         name: "{{ item.name }}"
@@ -387,6 +498,7 @@
         - { 'name': "{{ zeroclaw }}", 'status': "{{ zeroclaw_enabled   }}" }
         - { 'name': 'docker'        , 'status': true                       }
         - { 'name': 'open-webui'    , 'status': "{{ open_webui_enabled }}" }
+        - { 'name': 'swap-stopper'  , 'status': true                       }
 
     - name: Workstation | Linux | Software | LLMs | Services | Verify
       ansible.builtin.uri:
@@ -414,42 +526,11 @@
         - "{{ 'http://127.0.0.1:3000/health' if open_webui_enabled else dummy_website }}"
       ignore_errors: true
 
-    #- name: Workstation | Linux | Software | LLMs | Config | zeroclaw | Root
+    # Services - End #
-    #  ansible.builtin.blockinfile:
-    #    path: "{{ user_user.home }}/.zeroclaw/config.toml"
-    #    marker: "# {mark} ANSIBLE MANAGED BLOCK | zeroclaw | Defaults"
-    #    insertbefore: '(?m)^schema_version'
-    #    block: |
-    #      default_provider = "ollama"
-    #      default_model = "qwen3.5:4b"
-    #      provider_timeout_secs = 3600
-    #    state: present
-    #    create: false
-    #    backup: true
 
-    #- name: Workstation | Linux | Software | LLMs | Config | zeroclaw | Channels | header
+    # Models - Start #
-    #  ansible.builtin.lineinfile:
-    #    path: "{{ user_user.home }}/.zeroclaw/config.toml"
-    #    insertbefore: '(?m)^[channels]'
-    #    block: |
-    #      message_timeout_secs = 3600
-    #    state: present
-    #    create: false
-    #    backup: true
 
-    #- name: Workstation | Linux | Software | LLMs | Config | zeroclaw | Channels
+    - name: Workstation | Linux | Software | LLMs | Models | ollama | User Shell Functions
-    #  ansible.builtin.blockinfile:
-    #    path: "{{ user_user.home }}/.zeroclaw/config.toml"
-    #    marker: "# {mark} ANSIBLE MANAGED BLOCK | zeroclaw | Channels"
-    #    insertbefore: '(?m)^[channels]'
-    #    block: |
-    #      message_timeout_secs = 3600
-    #    state: present
-    #    create: false
-    #    backup: true
-
-
-    - name: Workstation | Linux | Software | LLMs | Models | ollama | Manual Script
       ansible.builtin.blockinfile:
         path: "{{ user_user.home }}/{{ item }}"
         marker: "# {mark} ANSIBLE MANAGED BLOCK | ollama | Manual Script"
@@ -459,17 +540,29 @@
             {% for llm in llm_installs %}
 
             echo -e "\n\n Starting {{ llm.model }} \n\n"
-            ollama pull {{ llm.model }} || {
+            ollama pull {{ llm.model }} ||
-              status=$?
+              status=$? &&
-              echo "ERROR: Pull failed with status $status, exiting."
+              echo "ERROR: Pull failed with status $status, exiting." &&
               return $status
-            }
+
             {% endfor %}
             echo "Done!!"
           }
           alias pull-llms="pull_llms "
           alias pull-models="pull_llms "
           alias pull-ai="pull_llms "
+
+          function ollama_stop {
+            for model in `ollama ps | tail -n +2 | awk '{print $1}'`; do
+              echo "Stopping ${model}..."
+              ollama stop $model
+            done
+          }
+          alias ollama-stop="ollama_stop"
+          alias ollama-kill="ollama_stop"
+          alias ollama-killall="ollama_stop"
+          alias ollama-kill-all="ollama_stop"
+          alias ollama-stop-all="ollama_stop"
         state: present
         create: false
         backup: true
@@ -535,7 +628,9 @@
 
     - name: Workstation | Linux | Software | LLMs | Models | ollama | Get Models
       ansible.builtin.shell:
-        cmd: ollama ls | tail -n +2 | awk '{print $1}'
+        cmd: |
+          set -o pipefail
+          ollama ls | tail -n +2 | awk '{print $1}'
       register: llm_models
       changed_when: llm_models.rc == 0
 
@@ -552,10 +647,15 @@
         block: |
           {% for llm in llm_installs %}
           {% if llm.model in llm_models.stdout %}
+          {% if "chat" in llm.purpose %}
+
           [providers.models.ollama.{{ llm.provider }}]
           model = "{{ llm.model }}"
           timeout_secs = 3600 # 1 hour, default is 120 (2 minutes), issue happens between 5 and 6 minutes though.
-          context_window = {{ llm_context }}
+          context_window = {{ [llm_context, llm.context, 16000] | min | int }}
+          max_tokens = {{ [llm_tokens, (llm.context / 4), 4000] | min | int }}
+
+          {% endif %}
           {% endif %}
           {% endfor %}
         state: present
@@ -572,25 +672,38 @@
           version: 1.0.0
           schema: v1
           models:
+
+          # | Role | `contextLength` | `maxTokens` | Notes |
+          # |------|-----------------|-------------|-------|
+          # | Chat / Edit  | 131072 (128K) | 16384 - 65536 | Qwen recommends 65,536 output for instruct models; use lower if VRAM is tight |
+          # | Autocomplete | 32768 - 65536 |  1024 -  4096 | Keep low for speed; autocomplete doesn't need long outputs                    |
+
             {% for llm in llm_installs %}
             {% if llm.model in llm_models.stdout %}
-            - name: {{ llm.provider }}
+            {% if "code" in llm.purpose %}
+
+            - name: {{ llm.provider }}_chat-edit
               provider: ollama
               model: {{ llm.model }}
               defaultCompletionOptions:
-                contextLength: {{ llm_context }}
+                contextLength: {{ [llm_context, llm.context, 24000] | min | int }}
-                maxTokens: {{ llm_context }}
+                maxTokens: {{ [llm_tokens, (llm.context / 4), 8000] | min | int }}
-            - name: {{ llm.provider }}-auto
+
+            - name: {{ llm.provider }}_autocomplete
               provider: ollama
               model: {{ llm.model }}
               defaultCompletionOptions:
-                contextLength: {{ llm_context }}
+                contextLength: {{ [llm_context, llm.context, 12000] | min | int }}
-                maxTokens: {{ llm_context }}
+                maxTokens: {{ [llm_tokens, (llm.context / 4), 2000] | min | int }}
               roles:
                 - autocomplete
+
+            {% endif %}
             {% endif %}
             {% endfor %}
         mode: "0644"
       when: codium_config
 
+    # Models - End #
+
   when: ansible_pkg_mgr == "pacman"

tasks/workstation/linux/software/lxqt.yml

@@ -17,4 +17,3 @@
       - "{{ lxqt_package }}"
       - "{{ sddm }}"
     state: present
-    update_cache: true

tasks/workstation/linux/software/tlp.yml

@@ -53,7 +53,6 @@
       - "{{ tlp_rdw }}"
       - "{{ tlp_smart }}"
       - "{{ tlp_run_script }}"
-    update_cache: true
 
 ## Configuration ##
 

tasks/workstation/shared/software/dwm.yml

@@ -17,7 +17,6 @@
       - "{{ make }}"
       - "{{ dmenu }}"
       - "{{ font_awesome }}"
-    update_cache: true
 
 
 ## Download ##

tasks/workstation/shared/software/i3.yml

@@ -66,7 +66,6 @@
       - "{{ i3_redshift }}"
       - "{{ i3_redshift_gui }}"
     state: present
-    update_cache: true
 
 - name: Workstation | Shared | Software | i3 | Install (yay)
   ansible.builtin.shell: "yay -Qs {{ item }} || yay --noconfirm -S {{ item }}"

test.yml

@@ -1,15 +1,16 @@
 ---
-# Harmonize systems rather than doing everything manually. :)
+# Used to single out plays which are either near the end of the playbook or are
+#   brand new and not safe to add to the full mix of things until finished.
 
-# Setup of any device connected to this repo.
+- name: Main | TEST Setup
-- name: Main | Default Setup
   hosts: localhost
   connection: local
   become: true
 
   tasks:
 
-    - ansible.builtin.include_tasks: tasks/header.yml
+    - name: TEST| Header
+      ansible.builtin.include_tasks: tasks/header.yml
 
     #######################
     #######################
@@ -17,7 +18,11 @@
     #######################
     #######################
 
-    - ansible.builtin.include_tasks: tasks/workstation/linux/software/llm.yml
+    - name: TEST | Main
+      block:
+
+        - name: LLMs
+          ansible.builtin.include_tasks: tasks/workstation/linux/software/llm.yml
 
     #######################
     #######################
@@ -25,4 +30,5 @@
     #######################
     #######################
 
-    - ansible.builtin.include_tasks: tasks/footer.yml
+    - name: TEST | Footer
+      ansible.builtin.include_tasks: tasks/footer.yml