Scheduled Tasks

AI scheduled tasks provide planned execution in the RESVRL AI autonomous cloud. Describe the work you want AI to perform repeatedly, and AI separates the work request from its Cron schedule so the platform can execute it at the specified time.

This is more than a reminder or a script timer. Each run starts an independent AI work session and produces traceable conversation history and results. Together with RESVRL infrastructure tools, scheduled tasks can support recurring resource checks, status summaries, troubleshooting assistance, and other authorized cloud-management work.

What scheduled tasks are good for

Scheduled tasks work best when the request is:

• Periodic and easy to describe in natural language;
• Verifiable through a message, resource state, or follow-up check;
• Low-risk enough to pause, edit, or delete when needed;
• Not dependent on a person providing input at every run.

Examples include:

• Summarizing the state of selected resources every day;
• Checking a service periodically and producing troubleshooting information when an anomaly is found;
• Reviewing resource usage or cost changes on a fixed cadence;
• Running an authorized infrastructure-maintenance workflow periodically.

Examples: turning recurring work into an autonomous workflow

The examples below come from the AI capability cards in the RESVRL console. They are requests that can be analyzed against the current account, resources, permissions, and available tools, rather than fixed scripts. Start with read-only checks before allowing any change operation.

Security and protection

• Anomaly baseline: compare login failures, port scans, and SSH brute-force trends over the last seven days, flag deviations, and produce a risk report.
• Security exposure scan: inspect public ports, security-group rules, and 0.0.0.0/0 exposure across VMs, compare with the previous day, and list new risks.
• VM protection scan: check snapshot, security-group, and Cloud Agent coverage for every VM, then identify missing controls and risk levels.
• Security posture delta: summarize new attack sources, attack types, CVEs, and auto-blocked IP changes, then explain rising indicators and suggest hardening steps.

Kubernetes and container operations

• Cluster change report: compare Kubernetes resource snapshots and report additions, deletions, and updates to Deployments, Services, ConfigMaps, Secrets, and RBAC bindings, with impact analysis.
• Container health trend: inspect Pod restarts, CPU, memory, error events, and recent deployments to identify applications that deviate from their operating baseline.
• Job failure monitor: inspect CronJobs and standalone Jobs, distinguishing causes such as OOM, image-pull failures, configuration errors, and timeouts.
• Disk capacity forecast: use the last seven days of VM disk growth to predict disks that may exceed 90% usage within 30 days, then recommend cleanup or expansion.

Spending and resource optimization

• Daily spend briefing: summarize daily and monthly spending by VM, disk, network, and container, and warn when the trend may exceed the budget.
• Monthly budget forecast: combine current-month spending with the last seven days’ daily average to forecast the month-end total and identify the largest contributing resources.
• Low-balance early warning: calculate the number of days the balance can sustain, estimate the depletion date, and check unpaid orders that could lead to service interruption.
• VM sizing optimizer: use CPU and memory peaks to identify consistently underutilized or overloaded machines and compare the cost impact of resizing.
• Snapshot coverage report: check each VM’s latest snapshot, retention count, and failed tasks, and identify machines with degraded protection.

When creating one of these tasks, put the scope, time window, thresholds, report format, and notification expectation in the work content. For example:

Every day at 09:00 UTC, inspect restart counts, CPU peaks, and memory peaks for all container applications over the previous 24 hours. Flag anomalies and generate a report; do not make changes.

These examples require access to the corresponding resources and platform tools. The thresholds above are illustrative and should be adapted to the service baseline. For actions such as suspending CronJobs, deleting resources, changing security groups, or resizing instances, ask AI for a plan first and keep a human approval step.

Creating a task

Describe the goal and frequency in AI Chat, for example:

Every day at 09:00 UTC, inspect my Kubernetes clusters and summarize abnormal workloads.

AI uses the task tools to create the schedule. A task has two parts:

• Work content: the request to complete during one run. Do not put scheduling words such as “daily” or “hourly” in this content;
• Cron schedule: when to run the task, expressed as a five-field Cron expression in UTC.

You can also view and manage existing tasks in the tasks workspace.

Execution rules

• Cron schedules use UTC;
• The minimum interval for an AI scheduled task is 5 minutes;
• The platform validates the Cron expression and work content when a task is created or updated;
• Tasks run asynchronously through a background queue and do not block the current Chat session;
• Every run creates a separate scheduled-task session and stores the AI messages and result;
• Execution requires an available AI model and sufficient account balance;
• The next execution time is calculated from the latest Cron schedule.

Status and results

The tasks workspace shows the schedule and execution status. After a run completes, open its associated scheduled-task session to review the AI processing and output.

Common states include:

• Pending: the task is created and waiting for its next scheduled time;
• Running: the background queue is processing the current run;
• Succeeded: the AI work completed and the next execution time was calculated;
• Failed: the run encountered an error; the platform records the reason and schedules the next planned run.

When a task fails, review the associated session and failure reason before changing the work content, adjusting the cadence, or deleting the task.

Managing tasks

AI can use the task tools to:

• Create a task;
• Paginate through the current user’s AI scheduled tasks;
• Update the Cron schedule or work content;
• Delete a task and its associated scheduled-task sessions.

Only the owning user can manage a task. Delete tasks that are no longer needed to avoid unnecessary model calls and cloud-resource operations.

How this fits AI autonomy

AI scheduled tasks provide one form of triggering in the autonomy loop:

1. Trigger: the Cron schedule reaches its execution time;
2. Understand: AI reads the saved work content and available context;
3. Act: AI processes the request through controlled platform tools;
4. Record: the platform stores the session, execution status, and timeline;
5. Continue: the task calculates its next execution time and waits for the next loop.

A scheduled task is not unlimited autonomous access. Start with low-risk, verifiable, and reversible work, and confirm the relevant resource permissions, tool capabilities, and organizational policies before enabling unattended execution.

Security and governance recommendations

• Do not use unattended, high-frequency tasks for irreversible or high-risk changes;
• Define resource scope, check conditions, and expected outcomes in the work content;
• Validate a task with read-only checks and status summaries before adding mutation actions;
• Keep audit records and a human-takeover path for critical tasks;
• Review the task list regularly and disable expired, duplicated, or no-longer-compliant tasks.

Related documentation

• AI Autonomous Cloud
• AI Chat Guide
• Virtual Machines
• Kubernetes