32dots HEIDELBERG AI
Curriculum › AnythingLLM course › anythingllm-03
Session 3 intermediate

Add AI agents & Community Hub skills

USE 0 - 20 min

Call a built-in agent and pull a skill from the Community Hub

Beyond document chat, AnythingLLM has built-in AI agents with customisable skills — small abilities the assistant can invoke to actually do things, not just answer. There is also a Community Hub for sharing and importing agent skills, system prompts, and slash commands. You invoke an agent inside a chat with the @agent directive in the message bar. This turns your private document workspace into something closer to an assistant that can take steps on your behalf, still under your control.

  1. 1 Open a workspace and type @agent at the start of a message, then a request — e.g. @agent summarise the newest document in this workspace and list its three key claims. The agent runs and reports back in the chat.
  2. 2 Browse the Community Hub from AnythingLLM's menu. It hosts shareable agent skills, system prompts, and slash commands contributed by the community.
  3. 3 Import one skill or slash command that fits your work (for example a summarising or extraction helper), following the Hub's import flow.
  4. 4 Set a workspace system prompt to shape behaviour — e.g. You are a careful research assistant. Always answer from the uploaded documents and cite them; say so when you are unsure.
  5. 5 Re-run a real task using the agent and your new system prompt, and compare it to a plain chat answer from lesson 01.
✓

You invoked an agent with `@agent`, imported at least one skill or slash command from the Community Hub, and set a workspace system prompt.

BUILD 20 - 30 min

Wire up a repeatable research task as an agent + system prompt

The payoff of agents and system prompts is repeatability: a task you do often becomes one short command instead of a paragraph of instructions each time.

Your task

Pick one task you repeat (summarise new uploads, extract methods, compare two papers), encode it with a system prompt and/or a Community Hub skill, and run it twice to confirm it is consistent.

  1. 1 Choose a recurring task from your real workflow.
  2. 2 Write a workspace system prompt that captures how you want it done (grounded, cited, specific format).
  3. 3 Add a matching Community Hub skill or slash command if one fits.
  4. 4 Run the task on two different inputs and check the outputs are consistent and grounded in your documents.
Deliverable

A workspace configured with a system prompt (and optional imported skill) plus two consistent runs of your repeatable task.