A mechanical hub with copper arms radiating outward, suggesting one agent coordinating many channels and tools.
Hermes is a hub, not a single tool: one persistent agent that reaches you across many channels and delegates work to many tools.

Hermes Agent is an open-source AI agent from Nous Research that runs as a persistent process on a server, a GPU box, or a serverless backend, rather than as a one-off command in your terminal. It keeps a memory that survives across sessions, it writes and improves its own skills as it works, and you reach it either from a terminal or from more than 20 messaging platforms through a single gateway. It is released under the MIT license and is written mostly in Python.

Hermes is not primarily a code editor. It is closer to a long-running assistant that orchestrates other tools, including the coding agents on this wiki. Its official skills include delegating a coding task to OpenCode or to OpenAI Codex, which it drives through its terminal tools. That places Hermes one layer above a tool like Claude Code : where a coding agent runs the read-plan-edit-run loop itself, Hermes decides what to work on, remembers the context, and hands the editing to a specialist.

Where Hermes sits in the stack

Gateway and platforms
Terminal (CLI, TUI) Telegram, Discord, Slack WhatsApp, Signal, Email, SMS One gateway process serves the CLI and 20+ messaging platforms
Agent core
Reasoning and tool-calling loop delegate_task subagents Spawns isolated subagents for parallel work, background tasks supported
Memory
Persistent memory Cross-session search (FTS5) User modeling (Honcho) Searches its own past conversations and builds a model of you
Skills
Procedural knowledge docs skill_manage tool Creates and edits its own skills; compatible with the agentskills.io standard
Tools and backends
MCP servers Web search, image, TTS, browser Backends: local, docker, ssh, singularity, modal, daytona
Models
Nous Portal, OpenRouter Anthropic, OpenAI, Google, Bedrock Local: Ollama, vLLM, SGLang, llama.cpp

Installation

Hermes installs with a single script on Linux, macOS, WSL2, and Android (Termux), or with one PowerShell command on native Windows.

bash
# Linux, macOS, WSL2, Android (Termux)
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
powershell
# Native Windows PowerShell
iex (irm https://hermes-agent.nousresearch.com/install.ps1)

After reloading your shell, log in and start the agent. The --portal flag sets Nous Portal as the model provider and enables the tool gateway.

bash
source ~/.bashrc          # or source ~/.zshrc
hermes setup --portal     # OAuth login and first-run configuration
hermes                    # start the CLI, or: hermes --tui

Two ways to use it

The first pattern is the messaging gateway. You configure a platform once, start the gateway, and then talk to the agent from your chat app of choice. The same slash commands work across every channel.

bash
hermes gateway setup      # interactive platform configuration
hermes gateway start      # start the gateway, then message the bot
# in any conversation, shared slash commands work:
#   /skills    list and run skills
#   /model     switch the model
#   /new       start a fresh conversation

The second pattern is delegation. Hermes carries a bundled skill for handing a coding task to OpenCode. It drives the OpenCode CLI through its terminal tool, so Hermes keeps the memory and context while OpenCode does the file edits.

text
terminal(command="opencode run 'Add retry logic to API calls and update tests'",
         workdir="~/project")

Skills can also be invoked directly and stacked. For example, /github-pr-workflow /test-driven-development fix issue #123 and open a PR chains two skills into one instruction.

Typical workflow

Step 1 Deploy Install Hermes on a VPS or serverless backend so it runs continuously.
→
Step 2 Connect channels Wire up the gateway to Telegram, Slack, email, or the terminal.
→
Step 3 Delegate and remember Ask for work; Hermes recalls context and delegates coding to OpenCode or Codex.
→
Step 4 Learn It writes new skills from the task and persists what it learned for next time.

How it compares

Hermes belongs in a different column from the terminal coding agents. It is a persistent, memory-backed orchestrator, not a stateless editor you launch per task.

Hermes AgentClaude CodeOpenCode
CategoryPersistent orchestrating agentCoding agentCoding agent
Runs asLong-running server processPer-session in your terminalPer-session, with a server
MemoryPersistent across sessionsPer-session contextPer-session context
Reach20+ chat platforms plus CLITerminal, VS Code, desktopTerminal, desktop, IDE, web
Does coding byDelegating to OpenCode or CodexEditing files itselfEditing files itself
Best forAn always-on assistant that coordinates toolsDeep Anthropic-native codingProvider-agnostic coding

When not to use it

  • You want an in-editor coding assistant. Hermes runs as a server process and delegates editing to other agents. For file-by-file work in your IDE, use a coding agent such as Cline or a terminal agent such as OpenCode directly.
  • You want the single best code editor. Hermes orchestrates specialists rather than being one. Deep coding still runs through OpenCode, Codex, or OpenHands .
  • You do not want to run infrastructure. The gateway means managing platform tokens and a VPS or serverless backend. A local per-session CLI has far less operational overhead.
  • You only need a quick one-shot task. The value of Hermes is persistence and memory over time. For a single edit, the setup is more than the job requires.

Further reading

Sources