Hermes Advanced Delegation: Model Routes and Profile Routes Explained

Hermes advanced delegation is a new official plugin from Nous Research that lets your Hermes Agent pick the right model, or the right profile, for every subagent it spawns. Out of the box, every subagent runs on one model. With Advanced Delegation you give Hermes a short menu instead — a fast route, a deep route, a local route, or a whole team of profiles — and it chooses per task. In this guide I’ll walk you through what it does, how to install it, how to set it up in two minutes and the gotchas to watch for.

What is Hermes advanced delegation?

Hermes already ships with subagents. When a job is big, the agent calls its delegate_task tool and spawns child agents that work in parallel with a fresh context, then hand back a summary. (I cover the basics in my guide to Hermes Agent subagents.)

The catch is that, by default, every one of those subagents runs on the same model. Either they all inherit your main model, or they all run on the single delegation.model you pinned in config. There’s no way to say “use a cheap model for this lookup and a strong one for that bug”.

The advanced-delegation plugin fixes that. It’s part of the Hermes Official Plugins from Nous Research (version 1.0.0, author NousResearch) and adds two extra modes next to the built-in default. Only one mode is active at a time.

Hermes core still handles 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?

Mode 1: model routes (model_map)

Model routes are a menu of models with a name and a one-line description each. The agent reads the names and descriptions and picks one per task. A typical setup has three:

• fast — cheap and quick: lookups, summaries, simple edits.
• deep — slow and careful: hard debugging and design work, with a higher reasoning effort.
• local — private: runs on your own machine through an OpenAI-compatible endpoint.

Here’s what that looks like in config.yaml, based on the official README:

delegation:
  backend: model_map
  backends:
    model_map:
      options:
        - name: fast
          description: "Cheap and quick: lookups, summaries, simple edits."
          provider: openrouter
          model: google/gemini-3-flash-preview
        - name: deep
          description: "Slow and careful: hard debugging and design work."
          provider: anthropic
          model: claude-opus-5
          reasoning_effort: high
        - name: local
          description: "Private: runs on this machine."
          base_url: http://localhost:8080/v1
          model: qwen3-coder
      default: fast

Each route needs a name, a description, a model and a provider or base_url. You can also set reasoning_effort, a service_tier (normal, fast or ultrafast), fallback_providers, api_key and api_mode. The README recommends provider: over api_key:, because an API key in config.yaml is stored in plain text, while provider credentials stay in .env or auth.json.

A task that doesn’t pick a route runs on the default you set. If you set no default, it runs on the parent’s own model.

Mode 2: profile routes (profile_routes)

Profile routes go a step further. Instead of a model, each route is one of your Hermes profiles. When the agent picks a profile, the subagent runs exactly as if you’d started hermes -p <profile> for that task — with that profile’s own model, credentials, skills and SOUL.md.

That means you can build a small team: a coder profile that writes and tests code, a researcher profile tuned for long web research, and so on. Your main agent hands each job to the right specialist.

delegation:
  backend: profile_routes
  backends:
    profile_routes:
      profiles:
        - name: coder
          description: Writes and refactors code; runs tests.
        - name: researcher
          description: Web research and long-form summaries.
      default: coder

The list is also an allowlist: profiles you don’t list can never be picked, and a listed name with no matching profile on disk is refused with a pointer to hermes profile list.

How to install the plugin

You need Hermes 0.22.5 or newer — the first release with pluggable delegation providers. Hermes 0.22.0 to 0.22.4 refuses to install or enable it. Run hermes update first if you’re on an older build.

• Command line: hermes plugins install advanced-delegation (installs the reviewed version from the Hermes plugin catalog).
• Desktop app: go to Settings → Model → Subagents → Install plugin. The button shows under the mode picker while the plugin is missing.
• Latest main branch: hermes plugins install NousResearch/hermes-official-plugins#advanced-delegation --enable.

Plugins are installed per profile, so install and configure it in the profile that does the delegating. The target profiles for profile routes don’t need anything installed — they just have to exist and have a working model.

Set it up with hermes delegation config

You don’t have to touch YAML. The plugin adds a guided setup:

hermes delegation config          # pick the mode, then edit its routes
hermes delegation config --show   # print the current mode and routes

• Pick the mode: Default, Model routes or Profile routes.
• Model routes opens a route editor built from hermes model’s own pickers: a name, a one-line description, a provider you have credentials for (or a custom endpoint), a model, an optional reasoning effort and a speed tier if the route offers one. Each route is checked before it’s saved.
• Profile routes is a checklist of your other profiles, an optional line about each, and a default.
• Default switches back to the built-in behaviour.

Everything is written to the profile’s config.yaml using Hermes’s own config writer, so your comments and other settings are kept. Changes apply to new sessions.

Switch modes in a chat with /delegation

Why does a switch normally wait for /new? Because the per-task choices the model sees are part of its tool list, and 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. --now is the opt-in when you want the new mode straight away — it costs one prompt-cache miss.

On Telegram, Discord and Slack, bare /delegation shows buttons; in the CLI, TUI and Desktop it uses the normal picker.

Fail-closed: no silent switches

One thing I like about this plugin is that it never quietly runs your work on the wrong model. If the selected backend is missing or disabled, delegate_task returns an error naming the backend and the fix, and no subagent starts. Inside a batch, a task that can’t be routed — an unknown option, a missing profile, a speed tier the route doesn’t offer, missing credentials — is refused on its own while the other tasks still run.

Every routed result also reports its route: the route name plus the provider, model, reasoning effort and speed tier the subagent actually ran on. So you can always check which model did which job. To go back without the plugin, run hermes config set delegation.backend default.

When should you use it?

• Cutting costs: send simple research and summaries to a fast, cheap route and keep the strong model for hard problems. Subagents usually burn most of a run’s tokens, so this is where the savings are.
• Privacy: keep sensitive tasks on a local route.
• Specialist teams: use profile routes so a coding profile, a research profile and a writing profile each do what they’re best at.
• Stay simple if you don’t need it: if one model is fine for every subagent, the built-in default is all you need.

For choosing the actual models, see my guide to the Hermes subagent model settings.

My verdict on Hermes advanced delegation

This is the missing piece for anyone running serious multi-agent work in Hermes. You keep one planner, and every subagent lands on the right model or the right specialist profile, with your keys never exposed to the model and no silent fallbacks. If you’re on Hermes 0.22.5 or newer, install Hermes advanced delegation, set up a fast and a deep route, and watch your subagent bills drop.

About Julian Goldie

I’m Julian Goldie — founder of Goldie Agency and the AI Profit Boardroom. I test new AI agents almost every day for 432K+ YouTube subscribers and run Hermes, OpenClaw, OpenCode, Claude and Codex side by side in my own Agent OS. Claim your free strategy session or join the AI Profit Boardroom.

FAQ: hermes advanced delegation

What is Hermes advanced delegation?

An official Nous Research plugin for Hermes Agent that adds two delegation modes: model routes, where the agent picks a named model per subagent, and profile routes, where a subagent runs as one of your Hermes profiles.

How do I install the advanced-delegation plugin?

Run hermes plugins install advanced-delegation, or in the Desktop app go to Settings, Model, Subagents, Install plugin. It needs Hermes 0.22.5 or newer.

What is the difference between model_map and profile_routes?

model_map gives the agent a list of named model routes such as fast, deep and local. profile_routes lets a subagent run as one of your Hermes profiles with its own model, credentials, skills and SOUL.md.

Can the agent see my API keys?

No. Only the route names and one-line descriptions reach the model. Provider, model, endpoint and key fields never become tool parameters.

Why does /delegation wait until /new?

Changing the mode changes the agent’s tool list, which is part of the prompt cache. Waiting for a new session avoids re-reading the whole chat at full price. Add –now to apply it immediately.

Related reading

• Hermes Agent Subagents
• Hermes Subagent Model
• Hermes Subagent Steering

Last updated October 2026. This page is a living guide to hermes advanced delegation — the facts here move fast and it is updated as they do.