Routines

Schedule agent queries on a recurring basis using cron expressions. Routines run analysis, generate reports, trigger workflows, and surface insights automatically — without manual intervention.

What Is a Routine?

A routine is a scheduled query that runs on a cron schedule. Each routine specifies:

• Schedule — A cron expression or a preset (hourly, daily, weekly, monthly).
• Query — The prompt sent to the routing pipeline.
• Target — A routing rule, workflow, or specific model.
• Output — Where the result goes (dashboard, webhook, email, or stored report).

Info: Routines are the automation layer of Xilos. Combine them with workflows, tools, and webhooks to build fully autonomous pipelines.

Creating a Routine

From the Dashboard

Open Routines

Navigate to Orchestrate → Routines.

Create new

Click New Routine.

Set the schedule

Choose a preset or enter a custom cron expression.

Define the query

Enter the prompt that the routine will execute.

Select a target

Choose a routing rule, workflow, or specific model.

Configure output

Choose where results are delivered: dashboard only, webhook, email, or saved report.

Save and enable

Click Save. The routine is active and will run on its next scheduled time.

From the API

cURL

curl -X POST https://api.xilos.ai/v1/routines \
  -H "Authorization: Bearer $XILOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "daily-cost-summary",
    "schedule": "0 9 * * *",
    "query": "Summarize yesterday'\''s LLM cost breakdown by model and identify the top 3 cost drivers.",
    "target": { "type": "workflow", "id": "wf_cost_analysis" },
    "output": { "type": "email", "recipients": ["finops@company.com"] }
  }'

Python

import xilos
 
client = xilos.Client(api_key="...")
 
routine = client.routines.create(
    name="daily-cost-summary",
    schedule="0 9 * * *",
    query="Summarize yesterday's LLM cost breakdown by model and identify the top 3 cost drivers.",
    target={"type": "workflow", "id": "wf_cost_analysis"},
    output={"type": "email", "recipients": ["finops@company.com"]},
)
 
print(routine.id)

Cron Schedules

Xilos uses standard 5-field cron expressions: minute hour day-of-month month day-of-week.

Presets

Custom Expressions

Warning: All times are in the timezone configured in your organization settings. Verify the timezone before relying on time-sensitive routines.

Manual Triggers

Run a routine on demand without waiting for its next scheduled execution:

• Dashboard — Open the routine and click Run Now.
• API — POST /v1/routines/{id}/trigger

curl -X POST https://api.xilos.ai/v1/routines/$ROUTINE_ID/trigger \
  -H "Authorization: Bearer $XILOS_API_KEY"

Manual triggers do not affect the routine's schedule. The next scheduled run proceeds as normal.

Routine Runner

The Routine Runner is the background process that executes scheduled routines. It:

1. Polls for routines whose next run time has passed.
2. Submits the routine's query to the specified target (routing rule or workflow).
3. Collects the response and routes it to the configured output.
4. Records the run in the routine's execution history.

Execution History

Each routine maintains a log of past runs:

• Timestamp — When the routine executed.
• Status — success, failed, or timeout.
• Duration — Wall-clock time for the run.
• Token usage — Input and output tokens consumed.
• Output preview — The first 500 characters of the response.

Failed runs are retried up to 3 times with exponential backoff before being marked as failed.

Output Options

Info: Combine the webhook output with Webhooks to pipe routine results into Slack, Microsoft Teams, or any HTTP-compatible system.

Use Cases

Daily Cost Analysis

Run a workflow every morning that summarizes the previous day's LLM spending, identifies cost drivers, and emails the report to the FinOps team.

{
  "name": "daily-cost-analysis",
  "schedule": "0 9 * * *",
  "query": "Summarize yesterday's LLM costs by model. Flag any model whose spend exceeded the 7-day average by more than 20%.",
  "target": { "type": "workflow", "id": "wf_cost_analysis" },
  "output": { "type": "email", "recipients": ["finops@company.com"] }
}

Weekly Security Review

Every Monday, run a query that reviews blocked queries from the past week and summarizes policy violations, attempted prompt injections, and restricted-topic access.

{
  "name": "weekly-security-review",
  "schedule": "0 9 * * 1",
  "query": "Review all blocked queries from the past 7 days. Categorize violations by type and severity. List any repeated offenders.",
  "target": { "type": "rule", "id": "security_analysis" },
  "output": { "type": "report" }
}

Automated Reports

Generate a monthly report combining query volume, cache hit rate, cost savings, and quality scores into a single summary delivered via webhook to a reporting system.

{
  "name": "monthly-platform-report",
  "schedule": "0 9 1 * *",
  "query": "Generate a monthly platform report: total queries, cache hit rate, total cost savings from caching and compression, average quality scores, and top 5 models by volume.",
  "target": { "type": "workflow", "id": "wf_monthly_report" },
  "output": { "type": "webhook", "url": "https://hooks.company.com/xilos-reports" }
}

Best Practices

• Start with presets — Use hourly, daily, weekly, or monthly presets before writing custom cron.
• Stagger schedules — Avoid running many routines at the same minute. Spread them out to balance load.
• Use workflows for complex analysis — Point routines at workflows when the analysis requires multiple stages.
• Monitor failures — Check execution history regularly. A silently failing routine can go unnoticed.
• Set timeouts — Configure a timeout per routine to prevent runaway queries from blocking the runner.
• Tag routines — Use tags to group routines by team or purpose for easier management.