Automation & Configuration
Schedule tasks, take direct actions, connect integrations, and configure your workspace.
Chat can act immediately or on a schedule. It can also help connect integrations, set environment variables, and manage MCP servers.
Scheduled Tasks
A scheduled task runs on a cron expression or at a set time. A prompt task is labeled Subagent and runs a saved instruction against the workspace at run time. A workflow task runs a named workflow. Open Automations in the workspace sidebar to inspect these background jobs alongside scheduled workflows, polling triggers, and webhooks.
The New Task button creates Subagents only. Configure workflow schedules, polling triggers, and webhooks in their workflows. Chat and MCP's steelengine_jobs tool continue to manage background jobs; they do not list every trigger shown in Automations.
Ask for tasks like:
- "Every morning at 8am, summarize new leads and post them to Slack"
- "Every Monday, pull last week's workflow errors and write a report"
- "Run the CRM sync workflow every six hours"
- "Pause the morning summary task"
- "Show me the last five runs of the weekly report task"
A task moves from created to scheduled to running, and ends paused, completed, or failed. Chat can pause, resume, complete, or delete a task, and can change its schedule or input while it is paused. Deleting a task asks for your approval first. SteelEngine rejects schedules below the workspace's minimum interval.
Monitoring Automations
The inventory shows the current workspace's automations that you have permission to view, ordered by recent activity. Use Active and Completed to switch views, then narrow the list with the type filter or search. Summary cards show separate enabled, paused, and finished counts; the cards do not change the filter.
The Horizon shows recent runs and upcoming scheduled starts in your browser's local timezone. Hover over a marker to see the grouped items, their provider and type, timestamps, and outcomes. Webhook events appear when triggered; they do not have a predictable next start.
Select a row to open its detail page. Subagents show their saved instructions and run history; workflow schedules show their cadence and executions; polling triggers show their interval and checkpoint; webhooks show their event configuration. Finished tasks without future runs show that nothing further is scheduled. Available run costs use the same credit conversion as Logs, and viewing execution details requires the appropriate log permissions.
Selecting Run now shows Starting… immediately, then Queued until execution begins. A running Subagent remains visible after a page refresh, and its detail page updates until the run finishes. Stopping future runs does not hide an execution that is still in progress.
Model and Context
Saved selections of retired models resolve to their current replacement before Chat-specific fallbacks are applied. When no model is saved, the preferred default is claude-opus-5 if it is available; otherwise the workspace uses its first available Chat model.
For a Sub Agent, choose the model and reasoning effort in the task's configuration. The choices follow the same model catalog, workspace restrictions, and supported reasoning levels as Chat. A task created from Chat inherits its selected model and attached context. Tasks with no saved model use the workspace's default Chat model. If a saved model loses its provider key or is blocked by the deployment, the run resolves to that default. Reasoning-effort, provider, and organization access checks still apply; fallback does not make a run possible when no provider is configured. See Model and Reasoning Effort.
Saved context keeps the resource IDs, selected workflow branch and blocks, and any captured view. Each run refreshes accessible resources and checks permissions again. The source conversation identifies where the task came from; later messages in that conversation do not silently rewrite the saved task.
Runs and History
A Sub Agent can make several model calls and tool calls during one scheduled occurrence. The complete occurrence counts as one run. Open the task's run history to inspect status, duration, final assistant output, available tool inputs and results, and failure details. A workflow launched by the Sub Agent also has its own workflow execution log.
The assistant receives the task prompt, completion condition when configured, run count, and the last ten job-history summaries. Every successful scheduled run records a summary with update_job_history and explicitly decides whether the saved task is complete. For an until-complete task, a completed decision requires complete_job to stop future occurrences; without a separate condition, fulfilling the saved instruction is sufficient. Unfinished and persistent tasks keep their schedules, subject to the run limit. The final run output and the job-history summary are separate records. Missing or failed history or required completion writes make the run fail, with available output and tool diagnostics retained.
New Subagent logs save model costs from reported token usage, including cached input rates. Run history shows the total in credits; log details show input and output costs and token counts, without a workflow base execution charge. Failed or cancelled runs retain usage reported by completed model steps. Older logs without saved cost do not have a breakdown.
Scheduled runs have no Approve / Deny prompt. Review the saved instructions and connected accounts before enabling the task. See Sub Agent for authorization and interruption behavior.
Direct Actions
For one-off actions, ask Chat to do the task directly:
| Request | What happens |
|---|---|
| "Send a Slack message to #eng that the deploy finished" | Posts to Slack using connected credentials |
| "Email the Q3 report to jane@example.com" | Sends through a connected email provider |
| "Create a GitHub issue for this bug" | Creates the issue in GitHub |
| "Call this webhook with the following JSON" | Makes the HTTP request |
Actions that write to an external system ask for your approval in the thread before they run. If an integration needs authentication, Chat should show the appropriate connect card or tell you which API key/environment variable is required.
Credentials and Environment Variables
Chat can help you connect OAuth integrations, inspect available credentials, and set environment variables:
- "Connect my Google account"
- "What credentials are connected to this workspace?"
- "Set the SCORING_API_URL environment variable"
- "Add OPENAI_API_KEY as a personal environment variable"
Listing environment variables returns metadata. Retrieving values requires the current user's separate value-read permission and records a best-effort audit event without values. Setting a value does not grant reveal access. These checks also apply when scheduled assistants use the environment tools. See secret access rules.
Connected workspace credentials are available to workflows and Chat actions according to workspace permissions.
MCP Servers
Chat can manage the external MCP servers connected to the workspace and the workflow MCP servers it publishes:
- "Add the Stripe MCP server using this API key"
- "Remove the old analytics MCP server"
- "List the MCP servers connected to this workspace"
Once configured, external MCP tools can be used from workflows and agent blocks. You create custom HTTP tools in Settings → Custom Tools, not from Chat.