Hermes Agent CaMeL

What Hermes Agent CaMeL does

This fork adds an opt-in CaMeL-style runtime guard to Hermes Agent for indirect prompt-injection experiments, described in a section titled Optional CaMeL Guard. Most of the README reproduces the upstream Hermes Agent documentation, so the fork keeps the feature set that text describes.

That feature set includes a terminal interface with multiline editing and slash-command autocomplete, a single gateway process for Telegram, Discord, Slack, WhatsApp, Signal and the CLI, and a learning loop in which skills are created from experience and past conversations are searchable with FTS5. It also has a built-in cron scheduler, isolated subagents, model switching with hermes model, and six terminal backends: local, Docker, SSH, Daytona, Singularity and Modal.

Key features

• Opt-in CaMeL-style runtime guard aimed at indirect prompt-injection experiments
• Hermes Agent terminal interface with multiline editing and slash-command autocomplete
• Single gateway process for Telegram, Discord, Slack, WhatsApp, Signal and the CLI
• Skills created from experience and FTS5 session search for cross-session recall
• Built-in cron scheduler and isolated subagents
• Six terminal backends: local, Docker, SSH, Daytona, Singularity and Modal

When to use it

• Experimenting with trust boundaries against indirect prompt injection in a Hermes Agent setup
• Comparing a guarded Hermes Agent fork against the upstream project
• Running a Hermes Agent gateway with an added security layer

Who it is for: Security-minded Hermes Agent users and researchers who want to test CaMeL-style trust boundaries.

How it fits with Hermes Agent

A fork of NousResearch/hermes-agent rather than a plugin; it keeps the Hermes Agent feature set and adds the CaMeL guard.

Note: It is a fork of NousResearch/hermes-agent whose README mostly repeats the upstream text, and the CaMeL guard is opt-in and described as being for prompt-injection experiments.

FAQ