Platform
Agent Studio

Build the right agent
for every job.

Team Assistant comes with a powerful general-purpose agent. Agent Studio lets you build specialized ones: scoped to exactly the right tools, knowledge, and permissions for a specific task or team.

Start FreeBook a Demo
Agent Studio canvas editor with node graph
Vaults
Agent Studio
Apps
Computer
Connections
Desktop
Marketplace
Scheduled
Security
Agent Studio
Capabilities

Why build custom agents.

Scope, specialize, and share agents built for exactly one purpose.

01

Specialized agents for recurring work.

A travel booking agent that only has access to the travel policy vault and the booking API. An onboarding agent that walks new hires through their first week using HR knowledge bases and IT provisioning tools. Purpose-built agents are easier to share across your team, simpler to schedule for autonomous runs, and focused on exactly one job.

Onboarding
Travel
Support
Reports
Billing
Research
02

Control exactly what each agent can access.

Every custom agent has its own scope: which connections it can use, which vaults it can read, which tools it can call. Build a high-privilege agent with direct infrastructure access for ops automation, then lock it down so it cannot reach anything else. Build a low-privilege agent for daily reports that only needs read access to two tools. The scope is the security boundary.

HR Agent
Travel Agent
03

Agents that run outside Team Assistant.

Custom agents are accessible via A2A (Agent-to-Agent) and AG-UI protocols. Embed a support agent in your own product. Let another AI system delegate tasks to your specialized agent. The agents you build in Studio are available wherever you need them.

04

Compose agents from sub-agents.

Build specialized agents and compose them into larger systems. A customer support agent delegates billing questions to a billing sub-agent and technical issues to a troubleshooting sub-agent. Each sub-agent has its own model, prompt, and tool set. Override instructions at the point of attachment without modifying the original.

TA
Team Assistant
HR Agent
Travel Agent
How It Works

From blank canvas to production.

1

Define the scope

What should this agent do? Which tools does it need? Which vaults should it read? Start with the purpose, then pick the building blocks.

2

Configure on the canvas

Drag connections, vaults, and sub-agents onto the visual canvas. Write the system prompt. Every configuration change auto-saves.

3

Test it live

Open a chat with your agent directly from the canvas. Iterate until it behaves the way you want.

4

Promote and share

Move to Staging for broader testing, then to Production. Publish to the Marketplace for your whole organization.

Under the hood

Built for technical teams.

Click to expand each section.

  • Three stages: Development, Staging, Production. Edit in Development, test in Staging, go live in Production.
  • Promotion syncs the full agent configuration: prompt, connections, vaults, sub-agents, and canvas layout.
  • Staging and Production are read-only. Changes only flow through promotion.
  • Configure which tools the agent can use and which credentials it authenticates with.
  • Share agents with your team: let others use them as-is or invite them to co-edit.
  • Publish to the Marketplace to make the agent available to your entire organization.
  • Built-in analytics: token usage, cost, active users, response times, and activity trends per agent.
  • Compare performance across Development, Staging, and Production stages.
  • Promote to Production when ready, roll back if something goes wrong.
In practice

Internal IT support, fully automated.

A mid-size company's IT team builds an "IT Helpdesk" agent in Agent Studio. The system prompt instructs the agent to triage support requests, look up solutions in the company's IT knowledge base, and create Jira tickets for unresolved issues.

Setup took 15 minutes: selected Claude Sonnet as the model, wrote triage rules into the system prompt, attached the IT Knowledge Base vault (read-only, 500+ guides), connected Jira for ticket creation (pre-approved), connected Gmail for incoming support emails, and added a Password Reset sub-agent for self-service walkthroughs.

The agent now handles 60% of incoming IT requests without human intervention. Average response time dropped from 4 hours to under 2 minutes. The IT team monitors token costs and interaction volume through the built-in analytics dashboard.

Explore more
Vaults
The knowledge source your agents learn from
Connections
Tools and APIs your agents can use
Computer
Where your agents execute their work