Automated multi-step workflows with cron scheduling, agent coordination, human-in-the-loop approval gates, and real-time run monitoring
Symphonies and Playbooks are OpenRails' workflow automation layer — the systems that turn interactive AI agents into autonomous business process engines that run on their own, on a schedule, without human intervention (unless you want it).
A Playbook is an automated multi-step task chain — a reusable sequence of agent calls, tool invocations, data transformations, conditional branches, and human approval gates. Anything an agent can do interactively, a playbook can do automatically and repeatably, with use-case-specific outputs.
A Symphony coordinates multiple agents working together in parallel on different aspects of a complex problem — think of it as multi-agent collaboration where each agent contributes its specialty.
Key Value: Playbooks let you "set and forget" complex AI workflows. Schedule a daily email triage, a weekly compliance check, a customer onboarding pipeline, or an automated incident response — all running autonomously with monitoring, alerting, and human-in-the-loop approval gates at critical decision points.
Playbooks are automated multi-step workflows that chain agent capabilities into repeatable processes. They go far beyond simple task sequences:
Schedule: Every morning at 8:00 AM
Check inbox → Classify each message by topic and urgency → Route to the correct team or person → Draft response suggestions for high-priority items → Send a summary digest to the manager
Trigger: New application received
Receive application → Verify submitted documents → Run compliance check (human approval gate) → Create account in downstream systems → Send personalized welcome package
Trigger: Alert received from monitoring system
Monitor alerts → Classify severity level → Pull relevant documentation and runbooks → Draft incident report → Notify on-call team via Teams → Track resolution
Schedule: Every Monday at 7:00 AM
Query databases for key metrics → Gather data from project management tools → Generate analysis and trend insights → Format into a polished report → Distribute to stakeholders via email
Schedule: Hourly
Check new data uploads → Validate format and completeness → Flag anomalies and outliers → Notify data team of issues → Generate data quality scorecard
Schedule: Weekly on Fridays
Scan new documents against regulatory rules → Flag potential violations → Cross-reference with policy database → Generate compliance summary → Route flagged items for human review
Both are powerful, but they serve different purposes:
Playbooks are automated task chains designed to run on a schedule or in response to a trigger. Each step executes in order, passing its output to the next. They are the backbone of repeatable, unattended automation.
Symphonies coordinate multiple agents working simultaneously on different aspects of a complex problem. Each agent brings its own specialty, and their combined outputs produce the final result.
| Aspect | Playbooks | Symphonies |
|---|---|---|
| Execution model | Sequential steps (with branching) | Parallel agents (with dependencies) |
| Scheduling | Cron schedules, triggers, on-demand | Primarily on-demand or triggered |
| Primary purpose | Automate repeatable multi-step processes | Coordinate multi-agent collaboration |
| Typical use | Email triage, onboarding, reporting, monitoring | Research projects, complex analysis, multi-perspective tasks |
| Human-in-the-loop | Approval gates at any step | Per-agent checkpoints |
Schedule playbooks to run on any pattern: every hour, daily at 6 AM, weekly on Mondays, or complex patterns like "first business day of each month." Timezone-aware scheduling ensures runs happen at the right local time.
Live dashboards show active runs, step-by-step progress, resource usage, and any errors. Historical run data enables performance analysis and optimization over time.
Get notified on run completion, failures, budget threshold breaches, or quality score drops. Alerts via email, Microsoft Teams, webhook, or in-platform notification center.
Configurable retry policies per step: immediate retry, exponential backoff, or failover to an alternative step. Failed runs can be resumed from the point of failure after the issue is resolved.
Insert mandatory human review at any step in a playbook. The playbook pauses and notifies the designated approver, then continues only after explicit approval. Critical for compliance, financial, and customer-facing workflows.
Playbook and Symphony configurations are version-controlled. Roll back to previous versions if a change causes issues. Compare versions to understand what changed and when.
| Step Type | Description | Example |
|---|---|---|
| Agent Call | Invoke an OpenRails agent with specific input | "Classify this email using the Triage agent" |
| Tool Invocation | Call a connected tool or integration directly | "Query the sales database for Q4 revenue" |
| Conditional Branch | Route execution based on previous step output | "If severity is critical, escalate to on-call team" |
| Data Transform | Map, filter, or restructure data between steps | "Extract customer names and email addresses from the results" |
| Human Approval | Pause for human review before continuing | "Compliance officer must approve before account creation" |
| Notification | Send an alert or message without pausing | "Post the daily summary to the team's Teams channel" |