Hermes AI Agent

Overview

Hermes is meant to become a useful operating layer for my projects and homelab: small-block task planning, safer code/documentation edits, project status memory, repair workflow notes, and eventually AI-assisted infrastructure workflows through Hermes LabOps.

Problem

Repair tickets can contain scattered information across customer notes, issue descriptions, payment details, internal notes, and workflow updates. The goal is to reduce friction by making ticket information easier to summarize, review, and act on.

My Role

I am designing the workflow, testing local and cloud model handoffs, organizing the project memory structure, and using small, reviewable task blocks so the system can help without making uncontrolled changes.

Tools Used

Current Focus

Hermes is moving toward a more reliable project-memory and local-AI workflow. The practical focus is making it useful without letting it sprawl: small edits, clear context, Git-backed memory updates, and human review before anything important changes.

• Use LM Studio and Qwen models for local experimentation.
• Coordinate local models and cloud models in multi-agent coding workflows.
• Keep an Obsidian memory vault backed by GitHub.
• Use small-block workflows for safe edits and memory updates.
• Support infrastructure notes, coding work, repair documentation, and project status.
• Keep human review before customer-facing or infrastructure-affecting actions.

Hermes examples must remain sanitized. Do not include customer names, ticket IDs, phone numbers, emails, API keys, tokens, private paths, internal URLs, internal infrastructure details, or logs.

Process

• Collect ticket data from RepairDesk-related workflows.
• Route structured data through Make.
• Send relevant data through a Cloudflare Tunnel bridge.
• Save local JSON and prompt files.
• Have Hermes summarize the issue, identify likely repair workflow, suggest next diagnostic steps, and draft customer-facing communication.
• Iterate on prompt structure and data formatting.

Challenges

• Keeping ticket data structured and consistent.
• Preventing sensitive data from appearing in public examples.
• Making the AI output useful without overstating certainty.
• Handling missing or incomplete ticket fields.
• Designing a workflow that can be expanded later.

Outcome

Built a working concept for AI-assisted repair ticket triage that connects real repair workflow tools with local AI automation. The project demonstrates automation, workflow design, technical troubleshooting, and customer communication.

What I Learned

• AI agents need structured context to be useful.
• Good automation depends on clean inputs and clear guardrails.
• Repair workflows benefit from consistent documentation.
• Customer communication should be clear, cautious, and evidence-based.

Hermes + RepairDesk Integration Roadmap

Hermes is designed as an AI-assisted workflow tool for repair operations. The goal is to help summarize tickets, identify missing information, draft customer-safe responses, and surface useful next steps while keeping a human in control of customer communication and repair decisions.

Hermes is designed to assist with summaries, drafts, and recommendations - not to automatically message customers, close tickets, change invoices, or make repair decisions without human review.

Future Improvements

• Add stronger validation for missing ticket fields.
• Improve customer-safe response generation.
• Add better logging and error handling.
• Create sanitized screenshots and diagrams.
• Build a more polished dashboard or review interface.

Sanitized Screenshots / Artifacts