Agent Deployments

A Deployment packages an Agent and Environment as a repeatable task. Run it on a five-field cron schedule, trigger it manually, or use both.

Beta

Agent Deployments is a Beta feature. Available runtimes may vary by environment.

Deployments and Endpoints are peer resources:

Resource	Trigger
Endpoint	Incoming REST, MCP, or SSE request
Deployment	Cron schedule or Run now

Create in the Dashboard

Open Console > Deployments, then select Create deployment.

Declarative create

The default Declarative create mode creates a managed Agent and managed Environment for the Deployment.

1. Enter a name.
2. Select an available runtime.
3. Optionally set an Agent description, system prompt, and Skills.
4. Enter the initial user message sent at the start of every run.
5. Configure execution limits. By default, there is no hard duration limit and inactive Runs stop after 5 minutes.
6. Set a five-field cron expression and IANA timezone, or enable Manual only.
7. Select Create deployment.

For example, a daily AI-news summary can use:

Setting	Value
Name	daily-ai-news
Initial message	Find today's important AI news, verify the sources, and write a concise Chinese summary with links.
Cron	0 9 * * *
Timezone	Asia/Shanghai

Advanced create

Select Advanced create to use an Agent and Environment already configured in your organization.

1. Select an available Agent.
2. Select a supported, non-archived Environment.
3. Enter the initial user message.
4. Configure execution limits.
5. Configure the schedule or choose Manual only.
6. Create the Deployment.

The selectors show eligible resources from Agents and Environments. Advanced create uses the selected Agent's current version; historical versions are not selectable.

Manage a Deployment

Open a Deployment to view Overview, Settings, and Runs.

• Run now starts a manual run. You may override the initial message and provide a conversation key or idempotency key.
• Pause stops scheduled triggers. Manual runs remain available while paused.
• Resume resumes scheduled runs.
• Archive permanently stops new runs.
• General settings update the name, initial message, and schedule for future runs.
• Execution limits configure an optional hard duration limit and a required inactivity limit for future Runs.
• Agent settings update a declarative managed Agent or rebind an advanced Deployment. Pause the Deployment first and wait for queued/running work to finish.

Overview summarizes the Agent, Environment, runtime, and schedule. Settings configures future runs. Runs shows status, timing, usage, errors, and available Sessions.

Inspect Runs and Sessions

The Runs tab shows whether each run was manual or scheduled, its status, latency, token usage, execution-limit snapshot, last activity, error summary, and creation time. Continued authenticated runtime activity keeps a Run alive past the inactivity limit; an optional hard limit always stops it at the configured duration.

When a Run reaches either limit, its status becomes Timeout. The Run indicates whether the cause was the hard duration limit or inactivity. SandBase terminates the Session and destroys its Sandbox after timeout.

When a Session is available for a run, select Session to inspect its conversation and outputs. Runs without a Session show Session unavailable.

Next steps

• Create a Deployment with the API
• Manage Deployments and Runs
• Working with Sessions