Summary
Sharbel A. walks through how to run Hermes Agent cost-effectively, structured around three tiers: free, affordable, and optimal. The core argument is that Hermes itself costs nothing — the expense comes from the AI models routed through it, and poor routing decisions are why many users overspend. The presenter opens with a concrete example: a client unknowingly paying $5,000 per month by pushing every task through premium models regardless of complexity.
The free tier requires either a local machine running a model via Ollama or reliance on rate-limited hosted free APIs — and the presenter notes that running capable local models typically demands a machine costing around $10,000, making “free” a relative term. The affordable setup pairs a $20/month Codex subscription with OpenRouter for lightweight tasks, while the optimal tier steps up to a $100/month Codex plan, which the presenter uses personally and considers sufficient for nearly all users. In both paid tiers, cheaper models like GLM 5.2 and Kimi K2 via OpenRouter handle summaries, cron jobs, and low-stakes tasks.
The video closes with a practical routing framework: use free or local models when privacy matters more than quality; use cheap OpenRouter models for anything where a bad answer costs only time; reserve Codex for coding, debugging, multi-step tool use, and production workflows. Notably, the presenter demonstrates asking Hermes itself to audit one’s workflow and suggest routing optimizations — a self-improving feedback loop that reduces ongoing costs without manual analysis.
📺 Source: Sharbel A. · Published July 22, 2026
🏷️ Format: Tutorial Demo
