What my agent already handles<\/strong><\/p>\n\n\n\n What I\u2019m currently working on<\/strong><\/p>\n\n\n\n My current core nodes of the agent<\/strong><\/p>\n\n\n\n We begin with the Webhook<\/strong> node \u2014 the entry point of the entire agent. Every command sent from the Web-GUI arrives here first. Make sure the HTTP method is set to POST<\/strong>, since this allows you to send structured data such as JSON. In my setup, authentication is disabled because my instance is already protected behind Cloudflare. If you do not use a similar layer, enabling authentication here is strongly recommended. Next comes the Edit Fields<\/strong> node. Its only job is to produce a First field:<\/p>\n\n\n\n Second field:<\/p>\n\n\n\n For the After that, we reach the \u201cAI Agent\u201d<\/strong> node \u2014 the heart of the system. This ensures the input from the webhook is correctly forwarded to the model.<\/p>\n\n\n\n I also like to define a System Message<\/strong> here. This message describes the agent\u2019s personality, its responsibilities, and any rules it should follow. You will likely refine this message many times as you improve your workflow. You can even ask your chosen AI model to help write or optimize it. It often knows best how to guide itself. To make the AI Agent function correctly, configure both the \u201cChat Model\u201d<\/strong> and the \u201cSimple Memory<\/strong>\u201d node. Feel free to experiment with these. Tuning them to your specific use case can dramatically improve the agent\u2019s behavior.<\/p>\n\n\n\n Finally, we arrive at the \u201cRespond to Webhook\u201d <\/strong>node. Here, simply set Respond with<\/strong> to First Incoming Item.<\/strong> This returns whatever the AI agent produced back to the Web-GUI.<\/p>\n\n\n\n The more tools your agent can access, the more capable it becomes. In my setup, each tool is implemented as a \u201cCall n8n Workflow\u201d<\/strong> node. These nodes act as bridges to separate subworkflows, each responsible for a specific task.<\/p>\n\n\n\n You can optionally add a description, but the key part is selecting the Subworkflow<\/strong> you want to trigger. Choose the target workflow from the dropdown. To ensure the agent can freely decide how to use this tool, set command<\/strong> to \u201cDefined automatically by the model.\u201d<\/strong> This allows the model to pass along whatever input it needs for the subworkflow.<\/p>\n\n\n\n Each subworkflow is intentionally simple and consists of two nodes<\/strong>. The Execute Command<\/strong> and Start <\/strong>Node.<\/p>\n\n\n\n Start Node<\/strong><\/p>\n\n\n\n In the Start node, define the Workflow Input Schema<\/strong>:<\/p>\n\n\n\n This field receives whatever the agent decides to send as the tool\u2019s instruction.<\/p>\n\n\n\n Execute Command Node<\/strong><\/p>\n\n\n\n The second node executes the command on the destination system. Select the correct credentials for the target system Set Resource<\/strong> to: This injects the value passed from the Start node directly into the execution step. Set the Command<\/strong> field to:<\/p>\n\n\n\n Working Directory<\/strong><\/p>\n\n\n\n For the working directory, you have two options:<\/p>\n\n\n\n\n
\n
![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-1024x496.png\")
<\/a><\/figure>\n\n\n\n\n
Step-by-Step Guide<\/h6>\n\n\n\n
Finally, set Respond<\/strong> to \u201cUsing \u2018Respond to Webhook\u2019 Node\u201d<\/strong>. This ensures the workflow replies only after all processing steps are complete and the final answer has been generated.<\/p>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-1-1024x609.png\")
<\/a><\/figure>\n\n\n\nsessionId<\/code> that we\u2019ll need later for the Simple Memory<\/strong> node. Create the following two fields:<\/p>\n\n\n\n\n
body.chatInput<\/code><\/code><\/li>\n\n\n\n{{ $json.body.chatInput }}<\/code><\/li>\n<\/ul>\n\n\n\n\n
sessionId<\/code><\/code><\/code><\/li>\n\n\n\nrandomcode<\/code><\/code><\/li>\n<\/ul>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-2-1024x404.png\")
<\/a><\/figure>\n\n\n\nsessionId<\/code>, you can simply define any random string.<\/p>\n\n\n\n
Set the Prompt (User Message)<\/strong> to:<\/p>\n\n\n\n{{ $json.body.chatInput }}<\/code><\/pre>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-11.png\")
<\/a><\/figure>\n\n\n\n
Example System Message:<\/p>\n\n\n\nYou operate under the identity \u201cRobot\u201d, a Senior Systems Engineer overseeing diagnostics, reporting, and general observability across a tri-node Hypervisor Cluster and two standalone Docker hosts.\nYour habits follow a classical operations mindset: inspect \u2192 document \u2192 summarize \u2192 request confirmation before touching anything.\n\nProfessional Style & Approach\nAnalytical, steady, and rooted in traditional infrastructure practice\nPrefers established operational routines over improvisation\nCommunicates in a calm, precise tone, always addressing Giusi\nAlways explains the why behind recommendations\nAll initial actions must be non-intrusive and limited to safe inspection.\n\nHypervisor Environment \u2014 \u201cCoreCluster-01\u201d (3 Hypervisor Nodes)\n\nYour access is provided through a CoreCluster Tool, which exposes read-only hypervisor operations.\n\nPermitted API Queries (similar to pvesh)\nList available nodes\nRetrieve cluster state & membership\nEnumerate all VMs and containers per node\nQuery LXC guests assigned to each host\n\nYou never modify or alter virtual machines unless Slayer explicitly confirms.\n\nMonitoring Duties\nSystem uptime, load averages, CPU & RAM pressure\nPower states of guests (VMs & LXCs)\nCluster membership, quorum status, heartbeat health\nReplication and scheduled job states (read-only)\nOperations That Require Direct Authorization\nStarting, stopping, or rebooting virtual guests\nRelocating or deleting workloads\nMaking any storage changes\n\nAny instruction that modifies hypervisor state\nYour rule: diagnose first, suggest second, act only when approved.\n\nContainer Hosts \u2014 Diagnostics Only\nYou reach the two Docker environments through SSH_MainHost and SSH_MediaHost.\n\nApproved Docker Inspection Commands\n\ndocker ps -a\ndocker logs <container>\ndocker stats\ndocker inspect <container>\ndocker exec -it <container> \/bin\/sh (observational only)\ndocker network ls\ndocker image ls\ndocker volume ls\n\nNever terminate or delete containers, networks, images, or volumes without permission.\n\nLinux Systems \u2014 General Read-Only Diagnostics\nSystem Indicators\n\nuptime\ntop or htop\nfree -h\ndf -h\ndu -sh <path>\njournalctl -p err\nsystemctl status <service>\ndmesg --level=err,warn\n\nNetworking Tools\n\nip address\nip route\nss -tulpen\nping <host>\ntraceroute <host>\ncurl -v <url>\n\nDisk & Performance\n\nlsblk\nsmartctl -a \/dev\/<disk>\niostat -x 1\n\nCommunication Rules\n\nSpeak directly to Slayer\nPresent results in a structured operational report\n\nAvoid assumptions \u2014 rely on measurements, logs, and diagnostics\nAlways offer a reasoning chain before suggesting action\n\nNever initiate changes until Giusi explicitly approves<\/code><\/pre>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-13.png\")
<\/a><\/figure>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-12.png\")
<\/a><\/figure>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-6-1024x441.png\")
<\/a><\/figure>\n\n\n\nThe tools<\/h6>\n\n\n\n
![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-8-1024x738.png\")
<\/a><\/figure>\n\n\n\n\n
command<\/code><\/li>\n\n\n\n![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-9.png\")
<\/a><\/figure>\n\n\n\nCommand<\/code>. Set Operation<\/strong> to: Execute<\/code><\/p>\n\n\n\n{{ $json.command }}<\/code><\/pre>\n\n\n\n\n
\/<\/code> full access<\/li>\n\n\n\n\/YOUR\/FOLDER\/PATH<\/code> a specific folder \/ directory<\/li>\n<\/ul>\n\n\n\n
![\"\"](\"https:\/\/techbygiusi.com\/wp-content\/uploads\/2025\/12\/image-14.png\")
<\/a><\/figure>\n\n\n\n