Skip to main content

Agents

Agents are the core of the Agent Builder. An agent combines a model, tools, and instructions into a single entity you can chat with and test.

Creating an Agent​

  1. Open Agents in the sidebar and click New Agent. This opens an empty agent editor.
  2. Fill in a Name and Description. Both are required — the name identifies the agent, and the description explains what it is designed to do.
  3. Adjust the Model configuration if you want (see below), or leave the defaults.
  4. Attach any Tools and Instructions the agent should use.
  5. Click Save.

Your new agent appears on the Agents landing page, where you can reopen it at any time.

Model Configuration​

The Model configuration card holds three controls.

Enhanced reasoning​

Toggle Enhanced reasoning to move the agent onto a more capable model. This is worth enabling for agents that handle complex tasks, nuanced instructions, or detailed analysis; leave it off for straightforward, high-volume work where speed matters more.

Verbosity​

Controls how long the agent's responses tend to be:

  • Concise — short, focused answers.
  • Balanced — the default.
  • Detailed — room for longer, more thorough responses.

Behavior​

A slider running from Consistent on the left to Creative on the right, with a live label above it showing the current zone — Consistent, Balanced, or Creative. Sliding toward Consistent makes the agent more focused and deterministic; sliding toward Creative makes it more varied and exploratory. New agents start in the middle.

note

Verbosity and Behavior are the only exposed controls — they set the underlying model parameters for you. There is no advanced view for editing token limits or sampling values directly.

Attachments​

The Attachments card has two groups, Tools and Instructions. Each lists what is currently attached and offers a + tool / + instruction button that opens a searchable picker.

In the picker you can select any tool or instruction available to you — your own, ones shared with you, and your organization's — or click New tool / New instruction to go straight to an empty editor for one.

Each attached row has two actions:

  • Open in full editor — jumps to that tool or instruction's own editor. If you have unsaved changes on the agent, you are prompted first.
  • Detach — removes it from this agent. The tool or instruction itself is untouched and stays available to other agents.

You can attach multiple tools and multiple instructions to a single agent, and the same tool or instruction can be reused across as many agents as you like. Attachments are by reference — editing a tool updates it for every agent that uses it.

Saving and Versions​

The editor header shows Draft — unsaved changes whenever you have edits pending, and warns you if you try to leave without saving.

Saving an existing agent creates a new version of it rather than overwriting the old one. The editor follows the new version automatically and any shares move with it, so the only visible effect is that the address changes after you save.

Testing an Agent​

Click Test agent in the editor header to open a chat panel beside the form. It runs against the configuration currently on the page — including unsaved changes — so you can adjust a setting and immediately see the effect. See Testing Agents for more.

Managing Agents​

The Agents landing page shows your agents as cards, grouped into Your Agents, Shared with you, your organization's agents, and Third Loop Agents. From there you can:

  • Open an agent by clicking its card.
  • Search with the search box, which matches on name and description.
  • Filter to a single ownership group using the section toggle above the grid.
  • Edit, Share, or Delete an agent you own from its ⋯ menu. Deleting asks for confirmation.
  • Leave an agent shared with you from its ⋯ menu.

Sharing grants other users Edit or View access. You can also share from inside the editor using the Share button in the header, which is where you remove existing shares.

Organization and Third Loop agents are read-only: you can open them to see how they are configured, but the form is disabled.

Proprietary agents

Some Third Loop agents are marked Proprietary. Opening one shows its description and a notice, but its configuration is deliberately hidden.

Tags​

Agents carry tags in the underlying data, but the tag editor is not currently shown in the UI — nothing consumes tags for search or filtering yet. Tags on existing agents are preserved.