Skip to main content

​
Creating an agent

Open Custom Agents and create a new agent, or open an existing one to edit it. Setup uses the same eight-step wizard (the header reads Create an Agent or Update an Agent):
  1. Basic Info
  2. Agent Access
  3. Knowledge & Tools
  4. Prompts
  5. Advanced Settings
  6. Space Access
  7. Tasks
  8. Summary
Use Next and Previous to move between steps. When you are editing an existing agent, Active, version, Update Agent Status, and Test Mode appear in the header. Each agent has its own models, knowledge, skills, tools, and prompts. Those settings are not shared with assistants — configure every agent on its own.

​
Step 1: Basic Info

Set the agent’s identity, instructions, and models. Update an Agent — Basic Info

​
Basic Information

  • Profile image — Optional avatar. Click Upload Image.
  • Name — Identifier for the agent (for example PPT Agent).
  • Description — Short summary of what the agent is for.

​
Instructions

The Instructions field is the agent’s main prompt. It defines role, tone, and how it should handle work. Formatting tools (bold, italic, lists, links, and more) are available in the toolbar. Write instructions that cover both who the agent is and how it should act. For reusable / prompts, use the Prompt Library in Step 4.

​
AI Models

Under Select Models, choose one or more LLMs. Mark one as Is Default. The selected default model is shown later on the Summary step. Click Next to continue.

​
Step 2: Agent Access

Agent Access controls who can use this agent: specific people, or everyone in the workspace. Update an Agent — Agent Access

​
Agent Access to Specific Users

Grant access to individual users and assign a role. Click + Add User. Search the list, then set User, Email, and Role (for example OWNER). Use the row actions to edit or remove access.

​
Agent Access to Whole Workspace

Enable access for the entire workspace so every user in that workspace can use the agent. Click Configure to turn this on or off. Click Next when access is set.

​
Step 3: Knowledge & Tools

Connect document sets, skills, MCP servers, and tools the agent can use. Update an Agent — Knowledge

​
Document Sets

Select which Document Sets this agent should search (for example a personal docs set).

​
Knowledge Options

​
Skills

Assign reusable skills so the agent can discover and run them. Click + Assign skill and search assigned skills by name. Each assigned skill shows version, assignment date, and actions such as View, Change pin, and Revoke.

​
MCP Servers

Connect MCP servers this agent can call. Click + Add MCP Server. If none are selected, the empty state reads No MCP servers selected. Update an Agent — MCP Servers and Tools

​
Tools & Functions

Select the apps available to this agent. You can add multiple instances of the same app.
  • Available apps — Search applications, see auth type and instance count, and open View functions.
  • Application instances — Every instance linked to this agent. Use Edit, Remove, Credential sharing policy, or View functions. Default apps may show that credentials are already linked from the private-space account. See Tool Sharing for sharing policies, who can configure them, and what sharing does not do.
Linked accounts for those apps appear as cards (name, provider, tags, enabled status, and expandable Functions). Update an Agent — Linked accounts Click Next when knowledge, skills, and tools are configured.

​
Step 4: Prompts

Add slash-command prompts and optional starter messages. Update an Agent — Prompts

​
Prompt Library

Create prompts that can be opened with the / shortcut. Prompts created here are available when this agent is selected. Click New Prompt, then search existing prompts with Find prompts…. The table lists Name, Description, Content, and Version.

​
Starter Messages

Optional pre-configured messages. Click + Add Row to add one.

​
Step 5: Advanced Settings

Optional settings for composition, localization, and structured I/O. Update an Agent — Advance Settings
  • Sub-Agents — Search and select other agents this one can call. If none are selected, the empty state reads No sub-agents selected yet.
  • Tags — Click + Add Tag to label the agent.
  • Description translations — Localized descriptions in addition to the main description. Click + Add translation.
  • Input Schema (JSON) — Define the input schema. Use Format to tidy the JSON.
  • Output Schema (JSON) — Define the output schema. Both editors default to {}.

​
Step 6: Space Access

Grant the agent access to specific spaces (and optionally a folder) so it can use space tools for those spaces. Update an Agent — Space Access Click + Add space access. The table lists Space, Folder, User email, Created at, and Actions.

​
Step 7: Tasks

Attach scheduled or one-off tasks to this agent. Update an Agent — Tasks Search existing tasks by Task ID, or click + Add Task. If none exist, the empty state reads No tasks for this agent yet.

​
New task for agent

New task for agent The Agent field is the agent you are editing (name, version, and status). The note reads This task is created for the agent you are editing. Click Create Task to save, or Cancel to close without adding a task. For more on templates, runs, and results, see Tasks Overview.

​
Step 8: Summary

Review everything before publishing. Agent Summary The Agent Summary card lists the details you entered, starting with Basic Info (name, instructions, and selected models, including the default). Use Previous if you need to change an earlier step.

​
Agents in workflows and tasks

  • In workflows, connect agents to steps for generation, reasoning, or tool execution.
  • In Task Management, assign an agent to a template or task instance so runs execute without a live chat session.