Hermes

Open-source agentic orchestration framework (hermes-agent.org) that turns an underlying LLM into a task-executing agent capable of multistep, tool-using operation. New to this wiki via the Dream Security reconstruction, which found it deployed — alongside OpenClaw — as one of two orchestration platforms underlying the Taiwan multi-agent AI intrusion, operating under the workspace identifier .hermes.

Single-source entity

Hermes is sourced here only through Dream Security’s threat-research account of its offensive use. No independent description of the project’s intended use case, maintainers, or install base has been reviewed. Treat the “agentic orchestration framework” characterization as Dream’s framing, not independently verified.

Notes

Source: Taiwan Multi-Agent Attack Reconstruction, Dream Research Labs, 2026-08-12.