This article is auto-synced from its in-app version in Tai.
Agents are AI assistants configured for a specific job — such as checking itineraries, answering customer care questions, or monitoring flights.
When you'd use this
Browse the Agents catalogue to find a built-in agent that fits your task, or create a custom agent tailored to your company's workflows. For example: configure an agent with your help articles attached as knowledge and your booking system as a connector, then make it available for your team to chat with.
Steps
Browse agents
In the Workbench sidebar, click Agents (under the AI Platform section).
The catalogue lists the agents available to you — built-in agents (managed by Nezasa), company agents shared across your team, and your own personal agents.
Use the search bar or the Ownership and Status filters to narrow the list. Your filter and sort selection is remembered, so it's applied again the next time you open the list.
Click an agent name to open its detail page.
Create an agent
In the Workbench sidebar, click Agents.
Click Create Agent in the top-right corner.
Fill in the General tab:
Agent ID — a unique identifier in
namespace/agent-nameformat (e.g.mycompany/helper).Name — the display name shown to users.
Description — what the agent does. This also helps the system decide when to use the agent.
Ownership — new agents default to Personal (private to you). Company Admins can choose Company to share it with everyone in your company.
Skills — select one or more skills to give the agent its capabilities.
Connectors — select connectors to give the agent access to external systems.
Knowledge — attach knowledge items so the agent can draw on your content.
Use the Behavior tab to write a system prompt and set a welcome message.
Use the Channels tab to choose how the agent can be used. Streaming API is the agent's chat switch, on for every new agent: turn it off and nobody can chat with the agent, while REST API calls and reports keep working. Enable REST API or Report for those channels; for REST API, see Setting up an agent's REST API. When Report is on with the HTML output format, optionally pick a Layout: Document for a report read top to bottom, Interactive page for one to scan and explore with tabs and filters. Leave it automatic and each run decides. An interactive page takes a little longer to generate than a document, since there's more of it to write.
Click Create Agent.
Edit an agent
Click the agent in the catalogue.
Update any field across the General, Behavior, LLM, Channels, or Permissions tabs.
Click Save changes.
Test an agent
Open the agent's detail page.
Click Test Chat (top-right) to open a chat session with this agent pre-selected.
If the agent's Minimum Workbench role for your company is above yours, Test Chat and the Test API tab don't appear. Ask a Company Admin if you need access.
If Test Chat is greyed out, chat is off for this agent: turn on Streaming API in the Channels tab and save. An agent with chat off doesn't appear in chat agent pickers, and a conversation already open with it stops at the next message.
Agent ownership
Every agent has an ownership type, shown in the Ownership column:
Built-in — managed by Nezasa and available to every company. Read-only (built-ins are Nezasa-owned and can't be copied).
Company — shared across your company. Everyone in your company can use it; only Company Admins can create, edit, or delete it.
Personal — a private agent owned by one person. You can create, edit, and delete your own personal agents.
To find agents by type, use the Ownership filter (which includes Personal for your own agents), or click My agents to show only your personal agents.
Release a personal agent to your company
A Company Admin can turn a personal agent into a shared company agent so the whole team can use it:
Open the personal agent's detail page.
Click Release to company in the top-right corner and confirm. If the agent has the REST API channel switched on, the confirmation also offers to let API keys and OAuth clients call it — ticked by default; untick it to keep the agent's accepted authentication as it is.
The agent becomes a Company agent. A Company Admin can reverse this later with Return to owner, which makes it personal to its original owner again — this ends any active chat sessions with the agent, except the owner's. API keys and OAuth clients can't call a personal agent, so no key can call it any longer; the confirmation dialog lists the active keys whose allowed agents include it. The agent's REST API then accepts Workbench user only.
Tailor a built-in agent for your company
Built-in agents are managed by Nezasa and shared across every company, but a Company Admin can adjust how a built-in behaves for their own company — without affecting other companies:
Open the built-in agent's detail page and go to the Permissions tab.
In the Your company's access section:
Turn Surface in TripBuilder off to hide the agent from your company's TripBuilder and SDK.
Optionally set a Minimum Workbench role to restrict who in your company may run it.
Click Save company settings. Use Reset to defaults to drop your changes and inherit Nezasa's defaults again.
The rest of a built-in agent stays read-only — these settings only change its access for your company. To review and adjust this across your whole catalogue at once, use the Agent access page.
Tips and limits
Browsing agents is open to all users. Any user can create, edit, and delete their own personal agents, and can copy any agent they can see into their own personal space; creating, editing, or deleting company agents — and releasing a personal agent to the company — requires the Company Admin role.
Company Admins see an extra Owner column listing who owns each personal agent.
Built-in agents are read-only and can't be copied (they're Nezasa-owned). Company Admins can still tailor a built-in's TripBuilder visibility and who may run it for their own company, from its Permissions tab.
To copy an agent, open its detail page and click Copy, then pick whether the copy is Personal or Company (Company Admins).
An agent can only access the skills, connectors, and knowledge items you explicitly assign to it.
Agents remember facts you ask them to keep — scoped to you and that one agent, never shared across agents or users. See Using chats.
Changes to an agent's skills or connectors take effect on the next chat session.
An interactive page is emailed as a PDF like any other report; its tabs and panels are opened out on paper. See Scheduling reports.
Related