Hermes delegation lets a business owner give one AI agent a big job and have it split the work across several helper agents that run side by side.
For a small team, that's the difference between one assistant doing five research jobs in a row and five helpers doing them at once.
The question for you isn't whether it's clever; it's whether it saves your team hours without adding risk.
Here's the business case, the three setups I'd use, and the risks to manage, with every detail taken from the official docs.
If you want help deciding where agents fit in your business, book a free strategy session with my team.
What Hermes Delegation Means For A Small Team
Think of your main Hermes agent as a manager.
When a job is big, it hands pieces to helpers called subagents, each with a clean desk and its own terminal.
Each helper reports back with a short summary, not the full pile of notes.
That keeps your main agent focused and stops one long job from crowding out the rest of the conversation.
The official delegation docs describe the tool behind this as delegate_task.
Your agent decides to use it on its own when a task is big enough; the docs say you don't need to ask it to delegate.
Where The Time Actually Gets Saved
The first saving is parallel work.
A single call can send a whole batch of tasks, and the helpers run at the same time instead of one after another.
The second saving is that you keep working.
Top-level delegations run in the background, so the agent hands you a handle at once and posts the result as a new message when it's done.
The third is focus.
Because only the final summary comes back, your main chat isn't buried under dozens of web searches and file reads.
Use Case: A Local Shop Comparing Suppliers
Picture a garden centre owner choosing between three compost suppliers.
Instead of one long research session, the agent sends three helpers, one per supplier.
Each looks at delivery areas, minimum orders and product range, then returns a short write-up.
The delegation patterns guide calls this "compare alternatives", and it points out the helpers can't influence each other's findings.
Your agent then compares the three summaries and gives you one recommendation.
The owner's job is to read one answer, not run three searches.
The same pattern works for comparing booking tools, card readers or delivery couriers.
Any decision with a few clear options splits neatly into one helper per option.
Use Case: A Service Firm Getting A Second Opinion
An accounting or legal practice lives on checked work.
Hermes has a /review command that starts a separate background helper whose only job is to review what the conversation just produced.
It looks at the last 10 messages as its starting evidence and has the normal helper tools, so it can open files and read them rather than guess.
Type /review on its own, or add a focus such as /review focus on figures.
The review lands back in the same chat, and your main agent can act on the findings.
It works in the CLI, the TUI, the Desktop app and every supported messaging platform.
🔥 Want to see how other owners split AI work across helpers?
Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS with Hermes set up, four coaching calls a week and 3,400+ members sharing the agent workflows they use in real businesses.
→ Join the AI Profit Boardroom
Use Case: An Agency Running Client Jobs Side By Side
A marketing agency might need the same audit run for several clients on a Monday morning.
One batch can carry one task per client, with each task's details written into its own context.
If your team wants the answers in a fixed format, each task can carry an output_schema so the helper returns structured data you can drop into a report.
When an answer doesn't fit the format, the agent gets one correction attempt, and the raw text is kept either way.
For agencies with very different jobs, the official advanced-delegation plugin lets each task pick a route: a fast option for simple lookups, a careful one for hard problems.
I've written about that plugin on its own in this guide to Hermes advanced delegation.
Keeping The Running Costs Sensible
Helpers are where most of the AI usage happens in a parallel run, according to the docs.
So the docs recommend a simple split: keep your main agent on a top model for planning, and put the helpers on a cheaper, faster one.
That's one setting in the config file, delegation.model, with an optional delegation.provider.
In the Desktop app, the same choice sits under Settings → Model → Subagents, and nothing changes until you press Apply.
The trade-off is that the default setup uses one helper model for the whole batch.
If some tasks need the stronger model, the plugin's named routes solve that.
The One Skill Your Team Needs: Writing A Full Brief
Helpers start with a blank slate.
The docs say they have zero knowledge of your conversation, earlier tool calls or anything you discussed before the hand-off.
So "follow up on that quote we talked about" gets you nothing useful.
A good brief names the supplier, the product, the deadline, the file location and what a finished answer looks like.
There's one shortcut for teams that work in a project folder: helpers automatically read the folder's AGENTS.md or CLAUDE.md file.
Put your house rules there once, such as tone of voice, file naming or which folders are off limits, and every helper follows them.
That's the cheapest training you'll ever do for an AI team.
How Many Helpers To Run At Once
The two official pages give different defaults here.
The delegation page says up to 10 helpers by default, while the configuration reference shows 3.
For a business, the safe move is to set the number yourself with delegation.max_concurrent_children in the config file.
Start low, around three, and raise it once you've seen the results and the usage.
If a batch asks for more helpers than the limit, Hermes returns an error rather than quietly dropping tasks, so nothing gets skipped without you knowing.
How You Get Hermes Delegation
Delegation is part of Hermes Agent itself, so there's no separate product to buy into.
The v0.22.6 release notes list the ways to run it: the CLI and installers, the Microsoft Store app, Docker and Hermes Cloud.
Update the CLI with hermes update.
The Desktop app installed from the .exe or .dmg installer gets this release with the next bundled release, so check before you plan around it.
The named-route options come from the official plugin, which its README says needs Hermes 0.22.5 or newer.
Risk 1: Helpers Can Borrow Another Profile's Access
The plugin's profile_routes mode runs a helper as one of your other Hermes profiles.
That's handy if you keep, say, a coding profile and a writing profile.
But the docs warn that listing a profile lets this profile's helpers use that profile's credentials, SOUL.md and terminal policy.
So only list profiles whose keys and terminal access you're happy to share.
For an agency, that means never routing one client's work through a profile that holds another client's logins.
Risk 2: Background Work Isn't Guaranteed To Finish
Helpers belong to the session that started them.
The docs say a Hermes restart doesn't resume a running helper, and the attempt is recorded as unknown.
Ctrl+C and the Stop button end only your turn, while the /stop command or closing the session ends the background helpers.
For anything that must run overnight or on a schedule, the docs point to cron jobs instead.
My Hermes cron job guide covers that for business tasks.
Risk 3: A Summary Isn't Proof
The patterns guide is direct on this: a helper's summary is just a summary.
If it says the job is done, check the file, the report or the figures before you send anything to a client.
Helpers also can't ask you questions, write to shared memory or send messages, so they can't contact anyone on your behalf.
How You See What Each Helper Did
Every hand-off writes a plain log file per task, so you can read exactly what a helper did, step by step.
The logs live in the Hermes cache folder and are cleared automatically after 7 days.
If a helper fails, Hermes says so in one clean line, and on Telegram, Slack or Discord that notice reaches the chat even when progress updates are switched off.
When the plugin routes a task, the result also names the route and the model it really ran on.
That gives an owner or manager a simple audit trail without reading code.
What Changed On 2 November
Hermes v0.22.6 came out on 2 November 2026, and one line in its notes matters here: "/delegation choices saved for the next session".
If your team uses the plugin, a choice such as /delegation default is now saved to the config file and takes effect when you start a new chat with /new.
Add --now to a switch and the current chat changes at its next turn, with one extra full re-read of the conversation.
That makes the setting predictable for a team: whatever you pick is still picked tomorrow.
📞 Not sure which jobs to hand to AI helpers first? Book a free SEO strategy session and my team will look at your workflows and show you where agents and SEO can save time. → Claim your free strategy session
Should Your Business Turn It On?
Use it if your week has research, review or audit jobs that split into clear, separate pieces.
Skip it for quick one-off questions, where calling the tool directly is faster.
Skip it for anything that needs a human answer halfway through, because helpers can't ask.
Start small: one batch of three tasks, written with full details, on a cheaper helper model.
Then read the results, check them, and scale up only what saved real time.
Want the full system with trainings and weekly calls? Join the AI Profit Boardroom.
Also On Our Network
- 🌐 The seven-step operator setup for delegate_task, worker models and saved backends
- 🌐 A nine-criterion scorecard for subagents, including where the docs disagree
- 🌐 Five copy-paste prompts that set up subagents inside your Agent OS
- 🌐 Which models to put in the planner and worker seats, with GoldieBench scores
- 🌐 The complete Hermes delegation guide on the AI Profit Boardroom blog
FAQ: Hermes Delegation For Business
Do I have to tell Hermes when to delegate?
No. The docs say the agent delegates on its own when a task is complex enough. You can still ask it to split a job if you want.
Can a helper agent email my clients?
No. Helpers are blocked from send_message, so they can't send messages across platforms. They also can't ask you questions or write to shared memory.
Is Hermes delegation a paid add-on?
Delegation is part of Hermes Agent. The named model routes and profile routes come from the official advanced-delegation plugin, which needs Hermes 0.22.5 or newer.
Is it safe to route work through another profile?
Only if you're happy to share that profile's keys and terminal access. The docs warn that profile routes let this profile's helpers use them.
Will a background job survive a restart?
No. A running helper isn't resumed after a Hermes restart. Use a cron job for work that has to finish no matter what.
📺 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 a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Pick one research or review job this week, write it as three clear tasks, and if you want a second pair of eyes on the plan, book a free strategy session and bring your Hermes delegation questions.











