LLM Provider
OpenRouter · Nous Portal · Custom Endpoint
↓
Hermes Core
ReAct Loop · Max 60 Iterations · Tool Dispatch
↓
Memory
Session · MEMORY.md (~800t) · USER.md (~500t)
Tools
40+ built-in · ML · Infra · Data · Cron
Skills
SKILL.md format · agentskills.io · MIT
↓
Sandbox Backend
Docker · E2B · Local · Modal · Fly.io
Financial Services
Fraud detection MLOps · vLLM latency diagnosis · W&B audit trail
hermes chat -q "diagnose vLLM latency spike on fraud-model-v3"
hermes chat -q "generate W&B governance report for Q4 model runs"
Healthcare & Life Sciences
Clinical RAG over 2M documents · FHIR structured output · CLIP multimodal
hermes chat -q "build RAG pipeline over clinical-notes/ with FHIR output"
hermes chat -q "run CLIP similarity search on radiology-images/"
Telecommunications ★
Network anomaly detection at 50K towers · SAELens interpretability · real-time alerting
hermes chat -q "detect anomalies in tower-metrics/ using SAELens"
hermes chat -q "schedule hourly network health report to Telegram"
Retail & E-Commerce
Black Friday 10x traffic scaling · Qdrant product search · demand forecasting
hermes chat -q "scale inference cluster for Black Friday load test"
hermes chat -q "index product-catalog/ in Qdrant with CLIP embeddings"
Public Sector & Defense
Document intelligence · 100K forms/day · Chroma policy search · air-gap support
hermes chat -q "extract structured data from forms/ at 100K/day throughput"
hermes chat -q "search policy-docs/ for FedRAMP ATO requirements"
Model-agnostic — switch LLM providers in one command. No vendor lock-in.
Skills compound — 1st deploy is manual; 10th is one command. Institutional knowledge captured.
Cross-platform ops — Telegram → TUI → Discord. Agent maintains context across surfaces.
Compliance-ready — 5 sandbox backends, session logging, skill quarantine, air-gap support.
MIT license — no licensing risk. Portable agentskills.io format works across editors.
Dual-use for training — same architecture powers Nous RL pipeline. Path to custom domain agents.