Playbooks
Save and replay proven extraction workflows
A playbook is a saved extraction workflow. It records exactly what the AI agent did during a successful run so future runs can replay it directly — without the agent needing to re-explore the page.
Why playbooks matter
When the agent runs for the first time against a website, it:
- Explores the page structure
- Tries multiple data sources (DOM, network responses, framework globals)
- Writes an extraction script
- Validates the output against your schema
- Retries if the output doesn't match
This process is thorough but takes time and LLM calls. A playbook captures the result — the navigation steps, the working extraction script, and the schema mapping — so subsequent runs skip steps 1–4 entirely.
| AI-driven run | Playbook run | |
|---|---|---|
| Navigation | Agent decides in real time | Replays saved steps |
| Extraction | Writes script from scratch | Reuses saved script |
| Validation loop | Yes (up to 12 iterations) | No — script runs deterministically |
| Speed | Slower | Fast |
| LLM usage | High | Minimal |
How a playbook is created
After a run completes successfully, click Save as Playbook. This captures the full execution — navigation actions, the extraction script, and the schema it maps to.
One playbook corresponds to one website + one schema combination.
Running a playbook
From a website's detail page, select the playbook and click Run. The agent follows the saved steps and executes the extraction script. If the page structure has changed enough that the script fails, the agent falls back to AI-driven recovery.
Editing a playbook
You can view and modify a playbook's recorded steps from the playbook detail page. Edit when:
- The website changed its layout and the extraction is breaking
- You want to adjust navigation (add a scroll, change a click target)
- You want to attach a different schema
Playbook status
A playbook can become stale if the target website changes. When a run using a playbook fails consistently, re-run the extraction in full AI mode and save a new playbook to replace it.