Assist is a platform where teams can build, host, and deploy internal AI software and agents in one place — giving companies a safe, controlled environment to standardize workflows and improve operations.

What are sandcastles?

Sandcastles are purpose-built internal apps created by non-technical teams on the Assist platform. They allow employees to build custom tools and workflows without engineering resources.

How much does Assist cost?

Assist uses pay-as-you-go pricing: provider costs + 5%, $0.17 per 1k tool calls, with no seat fees. Enterprise custom pricing is also available.

Do we need engineers to use Assist?

No. Assist is designed for non-technical teams to build and deploy AI tools. Engineers can use it too, but it doesn't require engineering resources to get started.

Scheduling a Subagent

A subagent on a schedule runs on its own at the time you specify, executes a saved prompt, and logs the result. This is how you turn a manual workflow ("every morning I open the support digest agent and ask it to run") into something the team can rely on without anyone remembering to trigger it.

Before you begin

The subagent you want to schedule must already exist. See Creating a subagent.
You need access to the AI chat.
The subagent should be able to do the work on its own. If it normally needs you to answer a clarifying question mid-run, refine the system prompt and the tool whitelist before scheduling — a scheduled run cannot ask you for input.

Steps

1. Open a chat with the agent

From the Subagents page, click the chat icon on the agent's card. This opens a conversation with that specific agent.

You can also schedule from the main chat by naming the agent, but doing it from inside the agent's own chat keeps things unambiguous.

2. Ask to schedule the prompt

Use a prompt that includes the four things a schedule needs: what to do, how often, what time, and which timezone.

"Schedule yourself to run every weekday at 8:00 AM America/New_York with the prompt: 'Pull yesterday's support tickets, group by severity, and post the summary to the #support-ops Slack channel.' Call the trigger 'Daily Support Digest'."

Other examples:

"Schedule a weekly run on Monday at 7:00 AM America/Chicago with the prompt 'Generate the weekly AR aging summary from NetSuite and email it to finance@company.com.' Call it 'Weekly AR Aging'."

"Schedule a one-time run for next Friday at 5:00 PM UTC with the prompt 'Run the end-of-quarter inventory snapshot and save it to the filesystem as /snapshots/q1-final.md.'"

The agent confirms the schedule it is about to create and asks you to approve.

3. Confirm the trigger

Approve the schedule. The agent creates a trigger on itself and tells you when the next run will be.

4. Verify the trigger appears on the agent

Go back to Subagents. The agent's card now shows a trigger indicator and lists the schedule (for example, "schedule (daily) — Daily Support Digest").

5. Watch the first run

When the scheduled time arrives, the trigger fires. To inspect what happened, open Agent Tasks in the left navigation (or go to /agent-tasks). Each scheduled run shows up there with:

Status: success, failure, or partial.
Duration.
A full execution log you can drill into.
Any tool calls the agent made.

If the run failed, open the log to see why. The most common causes are tools that need a credential the agent does not have access to, or a prompt that depended on something only present during your interactive session.

Managing schedules

You can ask the agent to manage its own triggers in chat:

"List your scheduled tasks."

"Disable the Daily Support Digest trigger."

"Change the Weekly AR Aging trigger to run at 6:00 AM Central instead of 7:00 AM."

These are first-class tool calls and the agent returns confirmation when each one completes.

Notifications

Scheduled runs can post their results to Slack. Failed runs send an immediate message; successful and partial runs are batched into hourly digests. To enable Slack notifications, your workspace must have Slack connected and your user profile must have a Slack identity connected. Talk to your workspace admin if this is not set up.

Notes and limitations

No mid-run prompts. A scheduled run cannot ask you a clarifying question. Build the prompt so the agent has everything it needs.
Time zones. Always specify a timezone in your scheduling prompt. If you omit one, the agent will guess based on workspace defaults, and it may not guess what you expect.
Repeated failures. A trigger that fails several times in a row may be flagged. Check Agent Tasks if you stop seeing runs you expected.
Idempotency is on you. If the prompt creates an entity (a ticket, a row, an email), make sure the prompt or the tools handle the case where the same trigger has already fired today. Scheduled runs do not deduplicate for you.