Agent Library
A saved agent is a helper you set up once and reuse: a name, a face, instructions on how it should behave, and a set of tools it's allowed to use. Chatting with a saved agent is different from a one-off chat because it always answers the same way, with the same context, every time you open it.
What you can do
| Feature | What it does |
|---|---|
| Your agents | One list of every agent you made, in every workspace you belong to, with what each one does and the tools it may use. Switching workspace does not change it. |
| New agent | Give it a name, instructions and the tools it may use. |
| Chat and Run | Chat opens a conversation with the agent. Run does its saved task once, if it has one. |
1. Open your Library
- From Agents, select Agent library, or go to
/agents/library. - Each agent shows its name, the workspace it belongs to, one line of what it does and its tools. An agent a workflow or a routine uses says so, for example "Seat in SEO" or "Runs Daily sweep".
2. Make one
- Select New agent.
- Give it a name, for example "Docs Editor".
- Write what it does, for example "You are the Docs Editor. Tighten guide copy for docs.devghouse.com: keep numbered steps in sync with the screenshot above them, cut filler words, and flag any step that no longer matches what is on screen."
- Tick the tools it may use. Only tools your workspace has turned on are offered.
- Optionally write a task for Run, and link a project so project tools act on it.
- Select Create agent. It is saved in the workspace you are in, so that workspace's routines and workflows can use it.
3. Chat, run, or edit it later
- Chat opens a conversation with the agent, with its instructions and tools already loaded.
- Run does its saved task once. With no saved task, it asks what to do this time.
- Editing an agent changes every future chat, and every routine or workflow seat that runs as it, from that point on. Past conversations stay as they were.
- Delete removes the agent. Its past chats stay readable; it just stops answering as itself.