Hermes Agent

Context

Combines conversation history, local context files, MEMORY.md and USER.md, skills, optional external memory providers, tool results, and configured MCP servers.

Actions

Runs terminal and browser tools, calls services, schedules jobs, sends messages, installs skills and plugins, and can operate through gateways or programmatic protocols.

Execution

Self-hosted on a local machine, server, container, or configured terminal backend; inference can come from selected remote providers or local endpoints.

Memory & rules

Built-in memory stores facts in local files; skills store procedures. Project instructions and installed plugins add more context and capability.

Permissions

Dangerous commands can require approval; channel access uses pairing or allowlists; container isolation and tool-by-platform settings provide additional boundaries. A `--yolo` option bypasses dangerous-command prompts.

Verification

Command output, logs, checkpoints, security audit commands, and user approvals support review, but self-hosting makes configuration and operational safety the operator's responsibility.

Best for

• A customizable assistant reachable across messaging channels
• Model-provider flexibility
• Users willing to operate an open-source agent stack

Limitations

• Editorial judgment: people wanting a zero-maintenance hosted assistant
• Editorial judgment: high-trust automation without time to harden gateways, tools, and plugins