Hermes Agent subagents are helper agents that your main Hermes agent spins up with a tool called delegate_task, so one request can be split into several jobs that run in parallel.
Hermes Agent is the free, open-source agent from Nous Research, and subagents come built in.
Each helper starts with a blank slate, does its piece of the work in the background and hands back a short summary.
I've made two videos on them, one on async subagents doing the work for you and one on controlling subagents live.
In this post I'll explain what Hermes Agent subagents are and how they work, then show the ways a business, agency or creator can actually put them to use.
Hermes Agent Subagents In One Table
Here is the business view before we get into the use cases.
| Question | Short answer |
|---|---|
| What are they? | They are helper agents spawned by the main Hermes agent with delegate_task. |
| Who makes them? | They are part of Hermes Agent, the free, open-source agent from Nous Research. |
| Do they run in parallel? | Yes, and the delegation docs say up to 10 run at once by default. |
| Do they block your chat? | No, because top-level delegations run in the background and post back when done. |
| What do they cost? | The feature is free, but each helper uses tokens from your model provider. |
| What can't they do? | They can't message anyone, ask you questions, write shared memory or schedule jobs. |
| Can you watch them? | Yes, with a live roster in the CLI, /agents in the TUI and a Subagents frame in Desktop. |
🔥 Want my exact How I Use Hermes Agent Subagents To Do The Work For Me setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
What Hermes Agent Subagents Are
Imagine your main agent as a team lead with a stack of independent jobs.
Instead of doing them one by one, it hires temporary helpers, writes each one a brief and lets them work at the same time.
Those helpers are subagents.
The docs describe them as child agent instances with isolated context, inherited tool access and their own terminal sessions.
Each one works independently, and only its final summary enters the main agent's context.
That keeps your main chat clean, which matters a lot on long jobs.
Top-level delegations also run in the background automatically, so Hermes hands back a handle straight away and you can keep working.
When a helper finishes, its result arrives as a new message in the same chat.
How Subagents Work, Step By Step
Here is what happens when your agent delegates.
- The main agent decides a job splits into independent parts, and the docs say it does this automatically when a task is complex enough.
- It calls
delegate_taskwith a goal and context for each part, either as one task or as a batch. - Hermes spawns the helpers, and each gets a fresh conversation, a terminal and the parent's tools.
- The helpers work in parallel while you carry on with the main agent.
- Each helper returns a structured summary of what it did, what it found and any files it changed.
- The main agent combines the results and reports back to you.
The one thing to understand is that helpers know nothing about your earlier conversation.
The docs put it bluntly: a subagent's only context comes from the goal and context fields the main agent writes.
So the quality of the brief decides the quality of the work.
Use Case 1: Parallel Content And Onboarding Builds
My first video on subagents covers real examples of parallel content and onboarding builds.
That is exactly the kind of job subagents suit, because each piece is independent.
One helper can draft a welcome email sequence while another builds an onboarding checklist and a third writes FAQ answers.
The main agent keeps the overall plan and stitches the pieces together at the end.
For a creator or a small business, that means a request that used to take several back-and-forth chats can come back as one finished bundle.
Use Case 2: Three Builds With Live Steering
My second video shows three builds I ran while watching the subagents live: a roadmap page, an SEO ROI calculator and an audit landing page.
The point of that update is that you can redirect a helper without killing the run.
The docs describe the controls: Ctrl+T or F6 opens the live roster in the classic CLI, /agents opens the tree in the TUI, and Desktop shows a Subagents frame with Steer and Stop.
There is one caveat I flag in the video and the docs back it up.
Steering is queued, so the helper reads your guidance at its next checkpoint rather than instantly.
My tip from that video is simple: start small and steer early, before a helper has gone too far down the wrong road.
🔥 Want my Hermes subagent workflows for your business? Inside the AI Profit Boardroom, I've got Hermes Agent training, the Agent OS and step-by-step walkthroughs on delegation and live control. You'll be learning alongside 3,400+ members with four coaching calls every week. → Get access here
Use Case 3: Parallel Research For An Agency
Agencies spend a lot of time researching competitors, keywords and markets.
The docs show a parallel research batch where three topics are researched side by side, each with its own focus.
For an SEO agency, that could be three competitors, each researched by its own helper with a clear list of what to check.
Each helper can return a structured answer, because delegate_task accepts an output_schema that the main agent validates.
If the answer doesn't match, the main agent sends one correction turn, and a miss after that still keeps the raw work instead of throwing it away.
That makes it much easier to drop results into a report or a spreadsheet.
Use Case 4: A Second Pair Of Eyes With /review
After your agent finishes a piece of work, typing /review sends an independent background subagent to check it.
The docs say the reviewer gets the last 10 messages as starting evidence and the full normal toolset, so it actually opens the files and runs the code.
When it finishes, its review lands back in the same chat so your main agent can fix what it found.
For anyone selling deliverables, that is a cheap quality-control step before work goes to a client.
Use Case 5: Splitting A Big Code Or Content Job
Large refactors and big content updates flood a single agent's context.
The docs show a multi-file refactor handed to one subagent with full instructions, so the main agent stays clean.
When several helpers edit the same project, delegation.worktree_isolation: true gives each one its own git branch so they don't overwrite each other.
The main agent then reviews or merges each branch.
Use Case 6: Long Campaigns That Report As They Go
Some jobs are long campaigns with many small, separate tasks, such as checking dozens of client pages.
By default, a batch sends back one combined message when every task has finished.
The docs describe an opt-in setting, delegation.independent_completions: true, that makes results land as each task finishes instead.
Tasks that share the same group name still come back together, which suits a set of results you want to compare side by side.
The docs say it's off by default because a 15-task call can become up to 15 separate wake-ups for the main agent.
So I'd only switch it on for campaigns where acting on each result straight away actually saves time.
What Subagents Won't Do For You
This matters for anyone worried about an agent going rogue with clients.
The docs say subagents can't use send_message, clarify, memory or cronjob.
So they can't message your clients, ask you questions, write to shared memory or schedule work in your name.
They also can't pick their own tools, because they inherit the main agent's enabled toolsets and the model can't widen them.
A helper that stalls for 450 seconds, or 1,200 seconds inside a tool, is interrupted and reported, so a stuck helper doesn't hold your session forever.
Keeping The Cost Under Control
Subagents are where most of a run's tokens go, according to Nous's delegation docs.
By default each helper uses the same model as your main agent.
You can pin a cheaper model for helpers with delegation.model, and I've written the full setup in my Hermes subagent model post.
If different jobs need different models, the official advanced-delegation plugin lets the agent pick a route per task, which I cover in Hermes advanced delegation.
You can also cap the number of tool-calling turns each helper takes with delegation.max_iterations, which defaults to 250.
How To Start This Week
- Make sure Hermes is up to date, because live subagent controls arrived in recent releases.
- Pick one job that splits cleanly, such as researching three competitors.
- Ask your agent to split it into parallel subagent tasks and give each one full context.
- Open the live roster and watch the first run, steering early if a helper drifts.
- Run
/reviewon the result before you use it. - Once it works, pin a cheaper subagent model to lower the cost.
If you'd like help working out where agents fit your SEO business, you can book a free SEO strategy session with my team.
🔥 Want the full Agent OS with Hermes subagents ready to go? The AI Profit Boardroom includes the installable Agent OS, Hermes training and four weekly coaching calls where you can ask me how to brief your subagents. → Join the AI Profit Boardroom
Related Reading
- Hermes subagent model explains how to cut costs by choosing the helper model.
- Hermes advanced delegation shows how to route jobs to different models or profiles.
- Hermes Agent sandbox covers keeping helper commands off your real machine.
Also On Our Network
- 🌐 The six-step Hermes Agent subagents operator setup
- 🌐 My feature-by-feature verdict on Hermes Agent subagents
- 🌐 How subagents fit an Agent OS workflow
- 🌐 The worker models behind Hermes subagents, scored
FAQ: Hermes Agent Subagents
What are Hermes Agent subagents?
They are helper agents the main Hermes agent spawns with delegate_task, and only their final summaries return to your chat.
Are Hermes Agent subagents free?
The feature is free because Hermes is open source, but each helper uses tokens from your model provider unless you run locally.
What can a business use Hermes subagents for?
Parallel research, building several assets at once, reviewing work with /review and splitting big jobs into independent parts are the main uses.
Can subagents talk to my clients or send messages?
No, because subagents can't use send_message, clarify, memory or cronjob.
Do subagents remember my earlier conversation?
No, because they start fresh and only know the brief the main agent writes, plus project files such as AGENTS.md.
If you run an agency and want a plan for putting agents on client work, you can book a free strategy session and we'll map it out.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link-building agency, Goldie Agency, with 70+ people in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery", which covers nine link building strategies.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Start with one job that splits cleanly, brief each helper properly and watch the first run live, and Hermes Agent subagents will start doing real work for you.
