Hermes Agent
hermes-agent is a provider-agnostic AI agent framework that runs the same core agent across terminal, messaging platforms, and IDE-style control surfaces. Its distinctive move is to treat an agent not as a single chat loop, but as a durable operating environment composed of tools, memory, skills, sessions, and background workers.
Core idea
Most agent frameworks focus on one layer:
- orchestration libraries such as langgraph or openai-agents-sdk define control flow
- protocol layers such as mcp-moc define tool interoperability
- hosted runtimes such as anthropic-managed-agents-model define a managed execution environment
Hermes instead bundles all of these into a user-facing substrate:
- Conversation surfaces — CLI, messaging gateway, and ACP/IDE integrations
- Execution substrate — terminal, file, browser, search, vision, messaging, and scheduling tools
- Context layer — skills, bounded persistent memory, session search, and personality files
- Durability layer — profiles, gateway daemons, cron jobs, background processes, and multi-agent coordination
The result is closer to an agent shell or personal agent operating system than to a narrow SDK.
What makes Hermes distinct
1. Skills as procedural memory
Hermes treats SKILL.md documents as executable procedures that can be loaded on demand. This makes operational knowledge portable, inspectable, and improvable without hardcoding every workflow into the base prompt. See agent-skills-index, hermes-skills-system, and hermes-bounded-memory.
2. Bounded memory plus deep recall
Hermes keeps a small always-on memory in prompt and pairs it with searchable session history. This splits "critical facts the agent should always remember" from "everything the agent once discussed." See hermes-bounded-memory.
3. Same agent across many interfaces
The same Hermes instance can answer in a terminal, a Telegram thread, or another gateway surface while preserving the same tools and long-term context. See hermes-gateway.
3.5. A real operator control plane
Hermes does not force runtime management through natural-language chat alone. It exposes a structured slash-command layer for session control, model switching, background runs, gateway administration, and runtime steering. See hermes-command-control-plane.
4. Durable background execution
Hermes is not limited to foreground chat. It can schedule work, spawn background tasks, run daemonized gateways, and maintain isolated profiles. This gives it a lifecycle closer to infrastructure than to a one-off assistant process. See hermes-cron, hermes-profiles, and hermes-kanban.
5. Multiple collaboration modes
Hermes separates several forms of multi-agent work that other systems often blur together:
- hermes-subagent-delegation for synchronous in-turn child agents
- hermes-cron for scheduled autonomous reruns
- hermes-kanban for durable cross-profile work queues
That separation makes the runtime easier to reason about than systems that treat every kind of concurrency as one generic "agent swarm."
Architectural consequence
Hermes collapses several patterns that are often separated in other stacks:
- tool calling
- memory management
- skill loading
- message transport
- background automation
- multi-profile isolation
This makes Hermes a strong reference implementation for the idea that an agent is not just a model-plus-tools loop, but a persistent context layer wrapped around interchangeable models.
When to start here
Start at hermes-agent when the question is:
- how one agent can persist across sessions and platforms
- how skills and memory interact in a real deployed system
- how chat surfaces, cron, and background daemons can share one substrate
- how to compare a full agent environment against narrower frameworks such as agent-development-kit, openai-swarm, or langgraph
See Also
- hermes-moc
- hermes-bounded-memory
- hermes-gateway
- hermes-profiles
- hermes-skills-system
- hermes-cron
- hermes-subagent-delegation
- hermes-kanban
- hermes-command-control-plane
- hermes-vs-adk-openai-agents-langgraph
- lit-hermes-architecture
- spec-hermes-agent-loop
- agent-knowledge-vault
- agent-skills-index
- memory-spectrum
- daemon-design-pattern
- agentic-frameworks-moc
- adk-moc
References
- Source:
C:\Users\executor\AppData\Local\hermes\skills\autonomous-ai-agents\hermes-agent\SKILL.md - Source:
C:\Users\executor\AppData\Local\hermes\hermes-agent\website\docs\user-guide\features\skills.md - Source:
C:\Users\executor\AppData\Local\hermes\hermes-agent\website\docs\user-guide\features\memory.md - Source:
C:\Users\executor\AppData\Local\hermes\hermes-agent\website\docs\user-guide\profiles.md - Source:
C:\Users\executor\AppData\Local\hermes\hermes-agent\website\docs\user-guide\features\cron.md