The best Hermes Agent tools are the terminal and file tools, then web search and the browser. Those give your agent hands, eyes and access to the live web. Everything else on this list builds on top of them.
📺 Watch: Hermes + Claude + Codex - The AI System That Does EVERYTHING!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
Below you'll get the ten tools worth switching on, ranked, plus how to turn tools on and off, which terminal backend to pick, and how to keep things safe. All of it comes from the Hermes docs.
What tools and toolsets are in Hermes
A tool is a single action the agent can take, like web_search or read_file. A toolset is a named group of tools you switch on together. Per the Hermes docs, the common toolsets are web, terminal, file, browser, vision, image_gen, skills, tts, memory, cronjob, code_execution, delegation and safe.
The point of toolsets is control. You don't have to give every agent every power. A research agent might only need web and browser. A coding agent needs terminal and file. Turn on what the job needs and leave the rest off.
If you want to learn how to set up agents with the right tools from day one, the AI Profit Boardroom has Hermes courses plus my Agent OS, where agents run with memory, skills and tools already wired together.
How we ranked the best Hermes Agent tools
- How much real work does it unlock? Tools that let the agent actually build, edit and fetch things rank first.
- How often will you use it? Daily tools beat once-a-month tools.
- Does it need extra setup? Tools that need separate credentials, like x_search, drop down the list.
1. Terminal and files
The terminal and file tools are terminal, process, read_file and patch. With these, Hermes can run commands, manage running processes, read files and make targeted edits. That's the difference between an agent that tells you what to do and one that does it.
This is also the toolset that needs the most thought about where it runs. More on terminal backends below.
2. Web search and extract
web_search finds pages and web_extract pulls the content out of them. Together they keep your agent's answers current instead of stuck at its training cutoff. For setup details, see my Hermes Agent web search guide.
3. Browser automation
The browser tools are browser_navigate, browser_snapshot and browser_vision. They let the agent move through real websites, capture what's on the page and look at it visually. Use them when web_extract isn't enough, like pages that need clicks or that load content as you go.
📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office
4. MCP servers
Hermes can use tools from MCP servers, which opens up a huge range of outside apps and services. If a tool you want isn't built in, there's a good chance an MCP server already has it. Here's how to connect one in my Hermes MCP server guide.
Plugin tools work the same way. Per the docs, there are about 500 community plugins in the Plugin Catalog, for example Home Assistant through the homeassistant catalog plugin. Browse them in the Hermes Agent plugins marketplace.
5. delegate_task (subagents)
delegate_task lets Hermes hand work to subagents. Big jobs get split into smaller ones. It sits in the orchestration group alongside todo, which tracks steps, and clarify, which lets the agent ask you a question when something's unclear instead of guessing.
6. cronjob
cronjob turns Hermes into something that works while you sleep. Per the docs it can create, list, update, pause, resume, run and remove scheduled jobs. Daily reports, weekly checks, morning summaries: set them once and let them run.
7. execute_code
execute_code lets the agent write and run code to solve a problem, like crunching numbers or transforming data. It's in the code_execution toolset. Pair it with a sandboxed backend if you're running anything you didn't write yourself.
📺 Watch: Hermes AI Agents Just Went Portable
8. Memory and session search
memory lets the agent add, replace and remove entries in its built-in memory. session_search lets it search past conversations. They're what make Hermes feel continuous from one day to the next.
9. Vision and image generation
vision_analyze lets the agent look at images. image_generate creates them. Handy for content work, screenshots and anything visual.
10. Text-to-speech and X search
text_to_speech turns text into audio. x_search searches X through xAI, but per the docs it's off by default and needs xAI credentials. Both are useful, just not daily essentials for most people.
Want help picking the right toolsets for your business? Book a free AI strategy session.
Ranking table
| # | Tool | Main tool names | Toolset | Use it for |
|---|---|---|---|---|
| 1 | Terminal and files | terminal, process, read_file, patch | terminal, file | Running commands and editing files |
| 2 | Web search and extract | web_search, web_extract | web | Current information |
| 3 | Browser automation | browser_navigate, browser_snapshot, browser_vision | browser | Interactive websites |
| 4 | MCP servers | MCP and plugin tools | Per server or plugin | Outside apps and services |
| 5 | Subagents | delegate_task | delegation | Splitting big jobs |
| 6 | Scheduled jobs | cronjob | cronjob | Recurring tasks |
| 7 | Code execution | execute_code | code_execution | Data and calculations |
| 8 | Memory and recall | memory, session_search | memory | Context across sessions |
| 9 | Vision and images | vision_analyze, image_generate | vision, image_gen | Visual work |
| 10 | Speech and X search | text_to_speech, x_search | tts | Audio and X research |
How to turn Hermes tools on and off
There are two ways, per the Hermes docs.
- hermes tools opens an interactive setup where you pick which tools are on for each platform. Use this for your everyday defaults.
- hermes chat --toolsets "web,terminal" sets the toolsets for one session. Use this when you want a tightly scoped agent for a single job.
My advice: start narrow. Turn on what the task needs, see how the agent behaves, then add more. An agent with fewer tools makes fewer surprising choices.
Terminal backends and sandboxing
The terminal backend decides where commands actually run. That matters a lot once your agent has terminal access.
| # | Backend | Where commands run | Notes from the docs |
|---|---|---|---|
| 1 | local | Your own machine | The default |
| 2 | docker | Isolated containers | Hardened: read-only root, capabilities dropped, PID limit |
| 3 | ssh | A remote server | Recommended so the agent can't modify its own code |
| 4 | singularity | Singularity containers | Container option |
| 5 | modal | Modal cloud | Remote sandbox |
| 6 | daytona | Daytona | Remote sandbox |
| 7 | vercel_sandbox | Vercel Sandbox | Remote sandbox |
If you're just testing on your laptop, local is fine. Once the agent is doing real work unattended, docker or ssh is the safer call.
The Tool Gateway
Per the docs, paid Nous Portal subscribers get web search, image generation, text-to-speech and browser automation through the Tool Gateway, without setting up separate API keys for each. If juggling keys is what's stopping you from turning those tools on, that's the shortcut.
Safety
- Never send passwords in chat on messaging platforms. The docs are clear on this one.
- Sandbox the terminal. Use the docker or ssh backend so commands can't touch more than they should.
- Use toolsets to limit scope. A research agent doesn't need terminal access.
- Keep credentialed tools off until needed. x_search is off by default for a reason.
Common mistakes
- Turning everything on. More tools isn't better. It just means more ways for the agent to go off track.
- Running unattended on local. Fine for testing. Not ideal for a cron job running overnight.
- Ignoring MCP. People hunt for a built-in tool when an MCP server already does the job.
- Skipping delegate_task. Long jobs crammed into one agent get messy. Split them.
I run Hermes and other agents with tools, memory and skills set up together inside my Agent OS. See how it fits together in the Agent OS guide, and how different setups stack up in Goldie Bench.
For the full Hermes courses and the Agent OS, join the AI Profit Boardroom for $69/mo.
FAQ
What are the best Hermes Agent tools?
The best Hermes Agent tools are the terminal and file tools, then web search and the browser. After that, MCP servers, delegate_task and cronjob add the most value.
How do I enable tools in Hermes?
Run hermes tools for an interactive per-platform setup, or start a session with hermes chat --toolsets "web,terminal" to pick toolsets for that session only.
Is it safe to give Hermes terminal access?
It's safer with a sandbox. Per the Hermes docs, the docker backend runs commands in hardened, isolated containers, and the ssh backend runs them on a remote server so the agent can't modify its own code.
Do I need API keys for every tool?
Some tools need credentials, like x_search via xAI. Paid Nous Portal subscribers get web search, image generation, text-to-speech and browser automation through the Tool Gateway without separate keys.
Want a plan for which tools your agents should actually use? Book a free AI strategy session and we'll pick the best Hermes Agent tools for your workflow.
