A recipe handles one task. A workflow chains tasks together. But some automations are bigger than a single workflow — they run for minutes or hours, coordinate multiple parallel workstreams, and need to survive failures gracefully.
Playbooks are JieGou’s answer to this. They execute workflows as async background agents with built-in checkpointing, crash recovery, and concurrency control.
What makes playbooks different
Standard workflow execution runs synchronously. The client waits for the result. This works well for workflows that complete in seconds to a few minutes. But for a 30-step automation that processes a batch of documents, enriches data from three external sources, waits for an approval, and generates a final report — synchronous execution hits practical limits.
Playbook execution is asynchronous. You start the playbook and get a run ID immediately. Execution continues in the background. You can check progress at any time, and the system notifies you when it completes.
Node-level checkpointing
The key architectural difference is checkpointing. After each step completes successfully, the playbook persists its state — all step outputs, the current position in the execution graph, and any loop or branch context.
If a playbook crashes mid-execution — a server restart, a transient infrastructure failure, a timeout on an external service — it doesn’t restart from scratch. It resumes from the last checkpoint, picking up where it left off. Steps that already completed are not re-executed.
This matters for long-running automations where re-running completed steps wastes time and money (LLM calls aren’t free) and can cause side effects (you don’t want to send the same email twice).
Concurrency control
Playbooks handle parallel branches safely. When a playbook reaches a parallel step, each branch executes independently with proper concurrency management. Resource locks prevent conflicting operations, and the playbook tracks which branches have completed to ensure the parallel step only finishes when all branches are done.
Progress streaming
While a playbook runs in the background, the UI streams real-time progress updates. You can see:
- Which step is currently executing
- Which steps have completed (with output previews)
- Which steps are pending
- Estimated time remaining based on historical execution data
This is the same progress view used for synchronous workflows, but it updates asynchronously via polling rather than a long-lived connection.
Retry and resume
When a playbook step fails, the system applies the standard retry strategy (exponential backoff, configurable max attempts). If retries are exhausted and the step fails permanently, the playbook pauses with an error status.
From the paused state, you have three options:
- Retry the failed step (after fixing the underlying issue, like a revoked API key)
- Skip the failed step and continue from the next one
- Cancel the playbook entirely
The ability to resume after a failure — rather than restarting from the beginning — is what makes playbooks practical for complex, multi-step automations.
When to use playbooks vs. workflows
Use standard workflows when:
- Execution completes in under 5 minutes
- You need the result immediately (synchronous response)
- The workflow is triggered by a user action and they’re waiting for output
Use playbooks when:
- Execution takes more than a few minutes
- The automation involves many steps or large data volumes
- Crash recovery matters (the cost of restarting is high)
- The automation runs in the background (scheduled, triggered, or fire-and-forget)
Sharing and analytics
Playbook runs support the same sharing model as workflow runs — private, department, account, or group visibility. Team members can view progress, inspect step outputs, and review completed runs based on their access level.
Playbook analytics show execution frequency, success rates, average duration, and cost per run. These help you identify which playbooks are most valuable and which need optimization.
Playbooks are available on Enterprise plans. Learn more about playbooks or contact us for Enterprise access.