Build & operate
Resources and storage
Choose models and stores, inspect persisted data, and understand ownership and retention.
Platform storage
Section titled “Platform storage”Resources includes platform MongoDB and file storage. For saved workflows, crawler-api resolves platform storage in the workspace owner’s scope. Use a Query node or the database/file browsers to inspect what a run actually saved; viewing an output is not the same as persisting it.
The product backend implements real read-only database and file routes. If a deployed view is empty, verify the selected store, workspace, run, permissions, and whether a storage node executed.
Bring your own resource
Section titled “Bring your own resource”Resources → Add Resource supports model endpoints and MongoDB resources. Select an authorized model in the LLM node or an authorized store in storage settings. Connection secrets are write-only in the product interface: saved canvases reference resources by id, while crawler-api supplies connections to the engine.
An unavailable resource may not be shared with you or may have failed its connection check. Inspect its status and choose a working resource before retrying.
What gets stored
Section titled “What gets stored”| Node | Persistent data |
|---|---|
| Data Source | raw_pages, pages, and llm_results for supplied artifacts; optional page images |
| MongoDB Store | Structured documents in records, with optional keyed matching |
| File storage | Referenced files in S3-compatible storage; output records carry references |
A Table, Chart, or Text node displays output; it is not a storage destination. Query can read the four supported document collections. Database browsing is read-only, with bounded paging and truncated previews; open a document to inspect its full content.
Ownership and lifecycle
Section titled “Ownership and lifecycle”Shared-workspace access does not make every other database accessible. The backend checks access before resolving a store or source run. Keep keys and bucket credentials out of prompts, node settings, examples, and exports.
Run summaries, events, intermediate outputs, and stored records have different lifetimes. The repository does not establish a single permanent retention guarantee or a universal “delete run and cascade every artifact” contract. If long-term retention or deletion is required, confirm the deployment’s policy and backup process with its operator. Keep explicit exports of results that must outlive transient execution data.