Hermes Agent, from Nous Research, is a serious open-source project. It runs on your own server, keeps memory across sessions, gets better at tasks the longer you use it, and you can reach it from just about any chat app. It's MIT-licensed and happily runs on a $5 VPS, so your data and your setup stay yours. For a developer who wants an agent that grows with them, it's a strong place to start.
Hermes is also hard to audit, and the reasons are baked into the design. It's a single agent that rewrites its own skills and remembers everything from past sessions, so it won't behave the same way twice, and nothing records why it made a given decision. It's built for a developer rather than a team, so there are no roles or approval steps, and the agent can read whatever secrets live in its environment. Self-hosting keeps your data yours. The guardrails are still yours to build.
How they compare
| General Input | Hermes Agent | |
|---|---|---|
| Self-hosted on your own infrastructure | ||
| Runs autonomously on a schedule | ||
| No-code builder for non-technical users | ||
| Human approval gates before sensitive actions | ||
| Complete, exportable audit log of every run | ||
| Every credential access recorded | ||
| Credentials encrypted, isolated from the AI model | ||
| Role-based access control for teams | ||
| Deterministic, reproducible runs | ||
| Per-execution cost transparency | ||
| Sandboxed, isolated execution |
When to use Hermes Agent
Hermes is a great pick for developers and tinkerers who want a fully self-hosted agent they can shape over time. If you like owning the whole stack, you want persistent memory and a self-improving skill set, and you don't mind wiring up your own guardrails, it gives you an open, MIT-licensed base to build on. It's also a solid foundation for generating training data and running RL experiments.
When to use General Input
General Input keeps the data sovereignty of self-hosting and adds the governance an organization needs.
- It does the same thing twice. Next to autonomous agents, General Input also runs plain code workflows that behave identically on every run, and it logs each one so you can prove it. An agent that rewrites its own behavior can't offer that.
- Approvals come built in. Sensitive steps wait for a person to sign off before they run, so you don't have to build the human-in-the-loop yourself.
- A real audit trail, not the agent's memory. Every run is recorded with its inputs, outputs, timing, and cost, and every use of a credential is logged. An auditor can read it, which is not the same as trusting an agent's private memory.
- Secrets stay out of the model. Credentials are encrypted and passed to a sandbox at run time, never into the model's context. Hermes hands its one agent whatever happens to be in the environment.
- Made for a team. Roles, shared credentials, and collaboration are there from the start, so the whole company can automate, not just the person who set up the server.
Better together
Use Hermes for personal projects and research where you want to own and tweak every layer. Reach for General Input when an automation has to be repeatable and shared across a team, while you keep the same say over where your data lives.