Hermes advanced delegation is advanced-delegation, an official Nous Research plugin that lets Hermes Agent pick a different route for every subagent it spawns. Out of the box, every delegate_task child runs on one model. With the plugin, the agent chooses per task: a named model route like fast or deep, or one of your Hermes profiles.
📺 Watch: Hermes Just Turned One AI Agent Into a Team
🔥 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
The plugin landed in the Hermes Official Plugins repo on 22 October 2026, and Nous Research announced it on X on 26 October. Here's what it does, the three modes, how to install it, and a few route setups you can copy.
What Hermes advanced delegation actually changes
Hermes already has subagents. The delegate_task tool spins up child agents with their own fresh conversation and terminal, and only their final summary comes back. If you want the full picture of how that works, read my guide to Hermes Agent subagents.
The problem was routing. With the built-in setup, you can point all subagents at one model with delegation.model and delegation.provider. Every child in a batch then runs on that same model. A quick lookup and a hard debugging job get the same brain.
Advanced delegation fixes that. Per the plugin README, it registers two delegation providers. Hermes core still owns the whole subagent lifecycle: spawning, isolation, depth and concurrency limits, background runs and results. The plugin only answers one question: which route does this task take?
The smart bit is what the model sees. Only the route names and the one-line descriptions you write reach the agent. The provider, model ID, endpoint and key never become tool parameters, so the agent can't invent a route. It picks from your menu or nothing.
I run Hermes every day inside my Agent OS with a profile per model, so routing work to the right profile is exactly the kind of thing this plugin is built for. If you want my full setup, the AI Profit Boardroom ($69/mo) gives you the Agent OS, a 30-day AI agent roadmap, daily tutorials as new tools drop, 4 coaching calls a week and the prompt library. 3,400+ members are in there.
The three delegation modes
You pick one mode at a time. Never two at once. The mode is stored as delegation.backend in your profile's config.yaml.
| Mode | delegation.backend | What the agent gets | Best for |
|---|---|---|---|
| Default | default | Every subagent on delegation.model and delegation.provider, otherwise the parent's model | Simple setups, one worker model |
| Model routes | model_map | A per-task option: one named model route from a list you define | Mixing cheap, strong and local models |
| Profile routes | profile_routes | A per-task profile: the subagent runs as one of your Hermes profiles | Specialist agents with their own skills and persona |
Default
This is built into Hermes and needs no plugin. Every subagent uses the delegation keys, or the parent's model when those are empty.
Model routes (model_map)
Each task item in delegate_task gains one extra property called option. It's a list of your route names in config order, and its description shows each name with its one-liner. Hermes refuses any other value before a subagent starts. A route resolves exactly like the delegation keys: provider uses the credentials you already set up, or base_url plus model targets an OpenAI-compatible endpoint.
Profile routes (profile_routes)
Here the extra property is profile. The subagent runs as that Hermes profile, with its own model, credentials, skills and SOUL.md. The README says it's exactly as if you'd started hermes -p with that profile for the task.
If a task picks no route, it runs on the mode's default when you've set one, otherwise on the parent's own model. And while model_map or profile_routes is active, delegation.model and delegation.provider are ignored. Switch back to Default and they apply again. Choosing those models is its own topic, covered in how to set the Hermes subagent model.
📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office
Version requirement and install
You need Hermes 0.22.5 or newer. The README calls it the first release with pluggable delegation providers. The plugin manifest declares that requirement, so Hermes 0.22.0 to 0.22.4 refuses to install or enable it. On cores older than 0.22.0 the plugin loads but does nothing and logs one warning. If you're still catching up on the 0.22 line, my Hermes Agent v0.22 breakdown covers it.
Install from the Hermes plugin catalog, which pins it to a reviewed commit:
- Terminal: hermes plugins install advanced-delegation
- Desktop: Settings, then Model, then Subagents, then Install plugin. The button shows under the mode picker while the plugin is missing.
- Follow main instead of the pin: hermes plugins install NousResearch/hermes-official-plugins#advanced-delegation --enable
One thing people miss: plugins install per profile. Install and configure it in the profile that does the delegating. The target profiles in profile routes need nothing installed. They just have to exist in hermes profile list with a working model.
Want help deciding which routes your business actually needs? Book a free AI strategy session.
📺 Watch: Hermes Bot Mode DESTROYS Grok Bot? (FREE!)
Setting it up with hermes delegation config
The plugin adds a setup command. Run hermes delegation config to pick the mode and edit its routes. Run hermes delegation config --show to print the current mode and routes without opening anything.
- Pick the mode. Default, Model routes or Profile routes. The cursor opens on the one you're using.
- Model routes: add a route. Give it a name (what the agent picks), a one-line description (what the agent reads), a provider you have credentials for or a custom endpoint, and a model from Hermes's own model picker. Reasoning effort is optional, and you can set a speed tier when the route offers one.
- Let it validate. Hermes checks the provider is known, the effort is a real level and the tier is one the route offers before saving.
- Set a default. When you hit Done with no default, it asks for one. You can choose No default, which sends unrouted tasks to the main agent's model.
- Profile routes instead? It's a checklist of your other profiles, an optional line about each one, and a default. Profiles keep their own models, so there's nothing else to pick.
Everything is written to the profile's config.yaml with Hermes's own config writer, so your comments and other settings stay put. Changes apply to new sessions.
Switching modes mid-chat with /delegation
Inside a chat, the /delegation slash command handles switching.
- /delegation shows the current mode, what each mode routes to and a picker.
- /delegation model_map saves that mode. It takes effect from the next session, after /new.
- /delegation profiles --now saves it and applies it to the current chat.
- /delegation default goes back to built-in behaviour from the next session.
Why wait for the next session? The tool list is part of the prompt cache. Changing it mid-chat makes the next request re-read the whole conversation at full price. The --now flag is your opt-in for that one cache break. On Telegram, Discord and Slack the picker shows as buttons. In the CLI, TUI and Desktop it uses the clarify picker.
Mid-chat safety is strict. If a task still picks a route from the old mode before /new, it's refused with a message naming --now and /new. It never runs on the wrong route. And if the configured backend can't run, say the plugin got disabled, Hermes refuses the subagent with an error pointing at /delegation.
Example route setups
These come straight from the plugin docs.
Model routes: fast, deep and local
Set delegation.backend to model_map, then list options under delegation.backends.model_map:
- fast: "Cheap and quick: lookups, summaries, simple edits." Provider openrouter, model google/gemini-3-flash-preview.
- deep: "Slow and careful: hard debugging and design work." Provider anthropic, model claude-opus-5, reasoning_effort high, service_tier fast.
- local: "Private: runs on this machine." base_url http://localhost:8080/v1, model qwen3-coder.
- default: fast, for tasks that pick nothing.
Each route needs a name, description and model, plus a provider or base_url. Optional keys are reasoning_effort, service_tier (normal, fast or ultrafast), fallback_providers, api_key and api_mode. The README says to prefer provider over api_key, because an api_key here sits in config.yaml in plain text.
Profile routes: coder and researcher
Set delegation.backend to profile_routes and list profiles under delegation.backends.profile_routes: coder ("Writes and refactors code; runs tests.") and researcher ("Web research and long-form summaries."), with default set to coder. That list is also an allowlist. Profiles you don't list can never be picked.
Big warning from the docs: listing a profile lets this profile's subagents use that profile's credentials, credential pool, SOUL.md and terminal policy. Only list profiles you're happy to share.
To see which models handle real agent work best before you build routes, check Goldie Bench. And the Agent OS guide shows how profiles fit together. For a hand mapping it out, grab a free AI strategy session.
FAQ
Is advanced delegation built into Hermes?
No. Default delegation is built in. Model routes and profile routes come from the official advanced-delegation plugin, which you install with hermes plugins install advanced-delegation.
Can I run model routes and profile routes together?
No. One mode is active at a time. Each keeps its own config block, so switching modes keeps the other's settings for later.
How do I know which route a subagent used?
Every routed result carries a route field with the route name plus the provider, model, reasoning effort and speed tier the subagent actually ran on.
How do I turn it off?
Run /delegation default. Without the plugin, the Hermes docs say hermes config set delegation.backend default restores the default.
Ready to give your subagents the right brain for every job? Join the AI Profit Boardroom for the Agent OS and step-by-step tutorials, or book a free AI strategy session and we'll plan your routes together.
