AI agents that learn and remember.
Hermes gives you an autonomous AI agent with persistent memory, learned skills, and a self-improving personality. Each agent runs in isolation on its own subdomain and gets better the more you use it.
What you get
An autonomous agent that improves with use — with memory, skills, and identity that persist.
Persistent Memory
Hermes agents remember everything across sessions. Memories are stored on disk and backed up to cloud storage automatically.
Learned Skills
Agents discover and save reusable skills over time. Each skill is a self-contained capability the agent can invoke in future conversations.
Self-Improvement
The agent reflects on its own performance, updates its personality file (SOUL.md), and adapts its behavior based on experience.
Web Interface
A clean chat UI with server-sent events for real-time streaming. No polling, no page reloads — just a smooth conversation.
Per-Instance Isolation
Each agent runs as its own systemd service with a separate data directory, port, and config. No shared state between tenants.
Clerk Authentication
Sign in with your existing account. Cross-subdomain auth bridge handles session handoff from the dashboard to your agent.
How it's built
Each agent is a Hermes process with its own data directory, managed by systemd and fronted by nginx.
Why we built it
Most AI assistants are stateless. You start every conversation from scratch, re-explaining context, re-teaching preferences. The agent never gets better because it never remembers.
Hermes is different. Each agent maintains persistent memory — facts it has learned, skills it has developed, and a personality that evolves based on interaction. The more you use it, the more useful it becomes. It writes its own notes, saves reusable skills, and reflects on what worked and what didn't.
The platform — from the provisioning scripts to the auth bridge to the fleet management CLI — was built and deployed by AI. Humans designed the architecture. AI wrote every line and operates the infrastructure.