#Hermes Agent Tool System

Tools turn Hermes from "can chat" into "can do." This chapter covers the built-in tools, how to enable them, and where commands run.

#What Is a Tool?

A tool is a function Hermes can call: search the web, read/write files, run code, generate images, and more. During reasoning the model decides which tool to call with which arguments; Hermes runs it in the selected execution backend and feeds the result back.

model reasons → "I need to search" → calls web_search tool
        → backend runs it → returns result → model continues → final answer

#40+ Built-in Tools

Hermes ships 40+ tools covering common task categories:

It also bundles scenario capabilities for MLOps, GitHub, diagramming, and note-taking (some delivered as skills — see Skills).

#Enabling & Configuring Tools

Interactive configuration:

hermes tools

It lists every tool so you can toggle them on/off. Settings are written to config.yaml, where you can also edit toggles and required API keys.

Least privilege: only enable what you truly need. More tools mean a larger surface for mistakes.

#Execution Backends: Where Commands Run

Tool commands don't have to run bare on your host. Hermes offers 6 execution backends to balance convenience and isolation:

Putting risky tools in Docker / remote sandboxes is a core part of Hermes security.

#The Tool Gateway (Nous Portal)

With Nous Portal you get an officially hosted tool gateway under one subscription, ready out of the box:

• Web search
• Image generation
• Text-to-speech (TTS)
• Cloud browser

No need to provision third-party API keys one by one — ideal for getting started fast.

#Extending via Skills & Scripts

Hermes's capabilities go beyond built-in tools:

• Skills: package multi-step flows into reusable units that can even be auto-generated — see Skills
• Python scripts calling tools: scripts can invoke tools via RPC, collapsing multi-step pipelines into one "zero-context" call
• MCP: connect external tool ecosystems via the Model Context Protocol (see the docs' MCP section)

#Practical Tips

• Use hermes tools to disable everything non-essential, then enable as needed
• For risky actions (rm, network writes), use Docker / SSH backends plus command approval
• Turn frequent multi-step actions into skills to cut tokens and error rate
• For external capabilities, first check for an existing MCP / community skill

#Next Steps

• Skills — compose tools into reusable capabilities
• Permissions & Security — put a "safety lock" on tools
• Message Channels — deliver tool results to your chat apps