Skip to content
MantleMantle docs

Agents and AI workers

Change how your assistant thinks and sounds on Settings > Agents, and the background jobs on Settings > AI workers.

For what agents, skills and tools are, see Agents, skills and tools.

Agents

An agent is something that holds a conversation. You talk to the Assistant. The others are specialists it hands work to, for example Researcher (web search), Pages (documents), Ledger (tables), Appsmith (apps) and Toolsmith (new tools). The Team Responder answers your members.

Edit an agent

  1. Open Settings > Agents and click an agent.
  2. Change what you need on its tabs:
Tab What you set there
General Name, description, avatar, role, priority, and its Telegram bot.
Model & routing Provider, API key and model; a backup route it falls back to; Voice (TTS); Thinking effort.
Behaviour System prompt, Tool groups, Skills, Delegates to.
Memory How much past conversation and knowledge each turn carries; Learned, which links to its notes in the Journal.
  1. Save.

Points worth knowing:

  • Backup route: an optional second model for when the primary is down. It can be a different provider, so a local model can run first with a cloud model behind it. Make backup primary swaps them.
  • Thinking effort: Inherit follows your profile. Off to Max override it for this agent. Higher costs more per turn.
  • Tool groups are the only way an agent gets tools. The Effective tools box shows the result. See Skills and tools.

Change many models at once

The Models tab shows every agent’s provider, model and key. Pick a new set for one row, and the others offer to copy it with one click. Nothing saves until you click Apply all. Only the primary route changes.

Add or delete an agent

Click New to add an agent. When you delete one, choose whether to keep or delete its chat.

AI workers

Workers are one-shot background jobs with no conversation: the Extractor reads new content, the Summarizer folds old chat into digests, the Reflector runs on a timer, and others handle documents, images, voice and image generation.

  • Each kind has one default worker. That is the one that runs.
  • Each worker has its own model and key. Chat-based workers, such as the Extractor and Summarizer, can also have a backup route.
  • Workers run every time content arrives, so they are where steady cost comes from. Check a model’s context window on Models before you switch the Extractor to it.

Provider keys belong on Settings > API keys. See Models and API keys. To run models on your own machine, see Local models.

Next