Deploy an Always-On Hosted Super Agent

Before You Start

Use a hosted team environment when a Super Agent needs to stay online for shared team work, hosted chat, tool calls, schedules, or external clients. Start from a Team workspace so ownership, billing, traces, and connected tools belong to the team instead of one person. A trace is the step-by-step run log for an agent request.

You need permission to manage the Team workspace, create or edit hosted agents, connect required tools, and review usage. Confirm the Super Agent has already completed the workflow manually before you make it always-on.

Decide the owner, allowed tools, expected runtime behavior, approval rules, and rollback path before deployment. Keep customer-facing actions, production records, and spending behind approval until the first runs are reviewed.

Choose or Create the Hosted Environment

Open the Team workspace that should own the hosted Super Agent. If your organization already has a ready hosted environment, deploy the Super Agent there so shared context, files, traces, and secrets stay under team ownership.

If no hosted environment exists, create one from the agent deployment or hosted environment setup area. Use a clear name that identifies the team or workflow, such as support-ops-hosted or revenue-automation-hosted.

Wait until the environment status is ready before adding production work. If you use the Operator CLI, run a read-only deployment status check to confirm you are connected to the intended organization.

Connect Tools and Context

Connect only the apps the hosted Super Agent needs for the first production workflow. For example, connect Gmail for inbox work, Slack for team updates, Linear for issue tracking, or Stripe for billing checks.

Attach the required tools to the Super Agent and add instructions for when each tool should be used. Include the source of truth, output format, and any approval rule for sending messages, changing records, or posting to channels.

Move durable team context into agent instructions, memory, files, or the connected system of record. Do not store passwords, raw secrets, or temporary task details in long-term memory.

Set Deployment and Runtime Settings

Enable hosted or always-on mode for the Super Agent, then confirm the slug, runtime, model/provider choice, and any channel settings such as Slack, Telegram, Model Context Protocol (MCP) clients, or public chat access. MCP is the standard that lets AI apps discover and call tools from another product.

Keep the first deployment narrow. Start with one workflow, one owner, and the smallest useful tool set, then add schedules, chat channels, or external MCP clients after the trace proves the hosted agent is behaving correctly.

Review credits and plan limits before enabling high-frequency schedules or public access. Hosted agents can consume credits through manual chats, scheduled jobs, external clients, and visitor traffic.

Monitor and Verify

Run a low-risk request against the hosted Super Agent after deployment. Confirm it reaches the right hosted environment, uses the expected tools, and either completes the task or asks for approval at the right step.

Open the latest trace and check tool calls, connected accounts, missing inputs, errors, and final output. If the agent has a public or channel endpoint, test from that endpoint in a fresh session too.

For the first week, review activity, usage, and failed runs daily. Pause or narrow the deployment if the agent uses unexpected tools, asks for repeated clarification, or spends more credits than planned.

Plan Safe Handoff and Ownership

Document the owner, backup owner, connected tools, approval rules, token storage location, and where teammates should check traces. Keep secret values in your secret manager rather than the handoff note.

Give teammates access through the Team workspace instead of sharing a personal account. If external clients or the Operator CLI use tokens, record who owns each token and when it should be rotated or revoked.

Before a teammate takes over, have them run a read-only check, inspect a trace, and explain how to pause the hosted agent or schedule. This proves the handoff covers operations, not only setup.

Troubleshooting

If the hosted Super Agent is not reachable, check that the hosted environment is ready, the agent is still hosted, and the slug or endpoint is correct. Retry after setup finishes.

If the agent cannot use a tool, reconnect the integration in the same Team workspace, enable the specific tool on the agent, and rerun a small read-only test.

If a schedule or external client reaches the wrong workspace, stop the run and confirm the active workspace, Operator CLI profile, MCP token, or channel connection belongs to the intended Team workspace.

If output quality changes after deployment, compare the hosted trace with the manual test trace. Update the instructions, memory, or runtime settings, then repeat the same verification prompt.

If usage is higher than expected, pause public access or recurring jobs, review plan limits, and narrow the prompt, tool list, or schedule cadence before turning the deployment back on.

Related Guides

Read "Connect Apps and Tools to Your Agents" before attaching production systems to the hosted Super Agent.

Read "Set Up the Operator CLI for Your Team" when teammates or automation runners need command-line access to the hosted team environment.

Read "Schedule Recurring Tasks and Automations" when the always-on agent should run on a cadence or app trigger.

Read "Deploy an Agent as a Public Chatbot" when the hosted agent also needs a visitor-facing URL.

Read "Read an Agent Trace" and "Manage Credits and Plan Limits" when you need to debug hosted runs or control usage.