Published: 2026-10-27
Summary

Hermes Advanced Delegation: send each subagent to a different model or profile with hermes delegation config

Chapters / key moments (click to jump — plays here on the page)

When a Hermes agent hands work to subagents, they normally all run on one model. The Advanced Delegation plugin lets the main agent choose a named model, or a whole Hermes profile, for each subagent. Julian Goldie's video explains it; because the video ends in a community pitch, the details here are checked against the plugin's catalog entry.

Source video

"New Hermes Delegation Update: AI Agent Teams!" by Julian Goldie SEO — Watch on YouTube → The video repeatedly promotes the presenter's paid community, which is not part of the plugin.

What the plugin does (official catalog entry)

From plugin-catalog/advanced-delegation.yaml in the Hermes Agent repository (maintainer NousResearch, tier "official", version 1.0.0, read 2026-10-27); docs are in hermes-official-plugins.

  • Two extra modes for delegate_task, alongside the built-in default. Model routes (delegation.backend: model_map): the agent picks a named model route per subagent from a list you define, such as fast, deep and local. Profile routes (delegation.backend: profile_routes): the agent runs a subagent as one of your Hermes profiles, with that profile's own model, credentials, skills and SOUL.md.
  • What the model sees. Only route names and their one-line descriptions reach the model, so write the descriptions as instructions for choosing.
  • Commands. hermes delegation config picks the mode and edits its routes, and --show prints the setup. The /delegation slash command saves a mode for the next session, or applies it to the current chat with --now.
  • Requirements. Hermes 0.22.5 or newer, the first release with pluggable delegation providers. Hermes 0.22.0 to 0.22.4 refuses to install it, and on a core older than 0.22.0 it installs, logs one warning and does nothing.
  • What it writes. Only delegation.backend and delegation.backends.* in the active profile's config.yaml. It makes no network or model calls of its own, and subagents run through core's delegate_task on the routes you configured.

Safety notes from the catalog

  • A model route's optional api_key is stored in config.yaml in plain text. Use provider: instead so keys stay in .env or auth.json.
  • Listing a profile under profile routes lets this profile's subagents use that profile's credentials, SOUL.md and terminal policy. List only profiles you are happy for your main agent to use.

How the video sets it up

  • Update to Hermes 0.22.5, then install the plugin: hermes plugins install advanced-delegation from a terminal, or Settings, Model, Subagents, Install plugin in Desktop. Plugins install per profile, so install it in the profile that does the delegating. (The catalog entry does not state the install command; check the plugin catalog docs.)
  • Run hermes delegation config, choose a mode, and for model routes use Add a route: a name such as fast, a one-line description, a provider you already have credentials for, a model, and optionally a reasoning effort and speed tier. Hermes validates each route before saving. For profile routes you tick profiles from a checklist and add a line about each.
  • His example routes: fast (cheap and quick, for lookups and summaries, on a Gemini Flash model through OpenRouter), deep (slow and careful, for hard debugging and design, on Claude Opus with high reasoning) and local (private, on a local model through an OpenAI-compatible endpoint). He also describes a default route, and says an invented route name is refused before a subagent starts.
  • Mode changes take effect from your next session (/new) because changing the tool list mid-chat can break the prompt cache; --now applies it at once. On Telegram, Discord and Slack he says buttons pick the mode.
  • His advice: start with model routes and two routes, set fast as the default, and watch which route each result took for a week before adding a local route or profile routes.

Gotchas & Caveats

  • The video says to run hermes delegation config show; the catalog entry spells it hermes delegation config --show.
  • The default route, the refusal of unknown routes, the prompt-cache reasoning and the chat buttons are the video's account; the catalog entry does not mention them.
  • The Hermes 0.22.5 release notes list pluggable delegation backends without detail, and say full curated notes arrive with v0.23.0.
  • We have not run the plugin. Whether the main agent picks sensible routes depends on your descriptions and the model driving it.

Key Takeaways

  • Delegation can now send cheap work to a cheap model: define a few named routes with one-line descriptions and let the main agent choose.
  • You need Hermes 0.22.5 or later, and routes with an api_key keep it in plain text.
  • Profile routes hand a subagent another profile's credentials, so treat the list as an allow list.