The Best Hermes Agent Tools, Ranked, And The Models That Use Them Well

The best Hermes Agent tools are the terminal toolset, web search and extract, and cron jobs, and that's the order I'd switch them on.

By tools I mean the built-in toolsets Hermes Agent can call, such as terminal, files, web, browser and cron, plus the companion tools I pair with it, such as MCP servers, Telegram, Obsidian, Ollama and Tailscale.

Hermes Agent is the free, open-source AI agent from Nous Research, and its tools are what let a model act instead of only answering.

This is my own ranking from using Hermes every day, so the order is my opinion and not a measured result.

Below you'll get the top 10, what each tool does, the real command that enables it, what it costs, and then a later section on which model to put behind them.

The best Hermes Agent tools at a glance

My top three are the terminal, web search and cron, because those three cover doing the work, knowing the facts and repeating the job.

What a tool is in Hermes Agent

The official docs say tools are functions that extend the agent's capabilities, and that they're organised into toolsets you can enable or disable per platform.

When I ran hermes tools on my own install, which is version 0.21.4, it listed 26 built-in toolsets, and 20 of them were switched on.

Every tool belongs to exactly one toolset, so switching on a toolset switches on everything inside it.

The file toolset is a good example, because it bundles read_file, write_file, patch and search_files.

Tools versus skills versus plugins

These three words get mixed up all the time, so here's the difference.

A plugin can deliver a new tool, and a skill can teach the agent when to use one.

I've ranked the other two in my lists of the best Hermes Agent skills and the best Hermes Agent plugins.

How to enable Hermes Agent tools

The whole system runs from one command.

Tools are set per platform, so a Telegram bot can carry fewer tools than your terminal.

The 10 best Hermes Agent tools, one by one

1. The terminal toolset

The terminal toolset holds terminal and process_manage.

You pick where commands run with terminal.backend in the config file, and the docs list local, docker, ssh, singularity, modal, daytona and vercel_sandbox.

It's best for building, installing and fixing things, and it's free on your own machine.

2. Web search and extract

The web toolset holds web_search and web_extract.

Firecrawl is the default backend, and a fresh install works with no key because Hermes rotates across the free tiers of Exa, Parallel, Firecrawl and Keenable.

Firecrawl's own free tier is 500 credits a month with a key, and SearXNG and DuckDuckGo are free search options.

It's best for research and fact-checking.

3. Cron jobs

The cronjob toolset exposes one tool, cronjob_manage, which creates, pauses, edits and removes jobs.

From the terminal, hermes cron create "every 2h" "Check server status" does the same thing.

Schedules accept forms like 30m and every 2h, and they accept cron expressions like 0 9 * * *.

A cron run can't create more cron jobs, which is a deliberate guard against runaway loops.

It's best for reports and monitoring, and the scheduler is free.

4. File operations

The file toolset is how Hermes edits a project instead of pasting code at you.

It's on by default, it's free, and it's best for coding and notes.

5. Subagent delegation

The delegation toolset holds delegate_task, which spawns child agents with their own context and terminal sessions.

The docs say up to 10 run at once by default, and only each child's final summary returns to the parent.

It's best for parallel research, and it multiplies your token use.

6. Browser automation

The browser toolset includes browser_navigate, browser_click, browser_type, browser_snapshot and browser_vision.

Pages reach the model as text-based accessibility trees, which suits AI agents.

You can attach your own Chrome with /browser connect or use a cloud provider such as Browser Use, Browserbase or Firecrawl.

It's best for sites with no API.

7. MCP servers

MCP lets Hermes use tools that live outside Hermes.

The command hermes mcp catalog listed 65 Nous-approved servers when I ran it on 10 October 2026, and hermes mcp install adds one by name.

For any other server you run hermes mcp add with --url or --command, and it's saved under mcp_servers in the config file.

It's best for apps such as Notion, Linear and Stripe.

8. The Telegram gateway

The messaging gateway turns Hermes into a bot on your phone.

You create a bot with @BotFather, run hermes gateway setup and paste in the token.

The docs say voice memos are transcribed automatically and scheduled results can be delivered to the chat.

9. Obsidian

Obsidian is a free notes app built on plain text files.

I point Hermes at my vault folder so it shares one memory with my other agents.

You have to give Hermes the full folder path, because it built inside its own workspace when I skipped that.

10. Ollama

Ollama runs models on your own computer, and ollama launch hermes --model qwen3.8 wired one into Hermes for me.

The Hermes docs also list LM Studio as a provider you can pick with hermes model.

It's best for private, offline work.

Other built-in toolsets worth switching on

The vision toolset analyses images you attach with /paste.

The image_gen toolset creates images through FAL.ai, and the tts toolset speaks with Edge TTS by default, which is free.

The computer_use toolset controls the desktop in the background on macOS, Windows and Linux, and it needs cua-driver from hermes computer-use install.

The memory and session_search toolsets let Hermes remember facts and search old conversations.

The code_execution toolset runs Python scripts that call Hermes tools.

The x_search toolset reads public X posts, but it needs xAI credentials.

Tailscale isn't a Hermes tool, but I use its free plan to reach my Hermes machine from my phone.

The Nous Tool Gateway comes with a paid Nous Portal subscription and routes web search, images, speech and cloud browsing through one account.

Limits and mistakes to avoid

The first mistake is switching everything on, because every tool is one more option the model weighs on each turn.

The second mistake is giving a messaging bot full terminal access without thinking about who can message it.

The third mistake is running ten subagents on an expensive model for a small job.

The fourth mistake is installing community MCP servers or plugins without reading what they do.

The honest limit is that a tool is only as good as the model deciding when to call it, which is what the next section is about.

Do Hermes Agent tools come with their own model?

No, they don't, because Hermes Agent is the body and you plug in the brain.

Every tool in this list is called by whichever model you've connected, so the same terminal tool behaves very differently under a strong brain and a weak one.

I use GoldieBench to choose that brain, and I want to be straight about what the board measures.

GoldieBench gives each model the same one-shot prompt to build a game, a page, a simulation or a visual, then scores the rendered result from 0 to 10.

It doesn't run a dedicated tool-calling test, so these numbers show how well a brain holds a hard multi-step job together, not how it scores on Hermes tools.

I still use it as a guide, because a model that finishes a hard build in one shot is usually a safer pair of hands on a long tool chain.

Which model to put behind your Hermes Agent tools

Every number matched the live model page when I checked on 10 October 2026.

How I match tools to brains

Light tools such as web search and file edits work fine on a cheaper model.

Scheduled jobs run many times, so I send them to a low-cost brain such as MiniMax M3.

Delegation and browser work involve many decisions in a row, so I move those up to Claude Opus 5 or Kimi K3.

Private jobs go to a local model through Ollama, and I keep the tool list short because a slower brain feels every extra tool.

Cron makes this easy, because a job can pin its own model with hermes cron create --model, separate from the model you chat with.

For the full ranking, read my post on the best LLMs for Hermes Agent, and for offline brains read the best local model for Hermes Agent.

A quick test before you trust a tool

I run the same routine every time I switch on something new.

First, I enable the tool in a spare profile so my main setup stays clean.

Second, I give the agent three jobs that should use the tool and three that shouldn't.

Third, I watch whether the model reaches for the tool at the right moments.

Fourth, I swap to a cheaper model and repeat the six jobs.

If the cheaper model handles it, I leave the tool there and save money.

Once the tools are working, my list of the best Hermes Agent use cases gives you real jobs to hand them.

Also On Our Network

• Agent Operatorsthe commands that enable every Hermes Agent tool
• AI Income Deskthe Hermes Agent tools a business should start with
• AI Tool Verdicttwelve Hermes Agent tools scored in one verdict
• agentos.guidewhere each Hermes tool fits inside an Agent OS
• agentos.guideJulian's Hermes and Obsidian second brain guide

Switch on the terminal, web search and cron, put a proven brain behind them, and you'll get the most from the best Hermes Agent tools.