Destination connectors
Destination connectors define where smart bites and embeddings go after extraction and (optionally) embedding. Send output to the Vector Catalog for search and agents, or to your own object storage or vector store for custom pipelines.
Vector Catalog
- Bundata Vector Catalog — The default destination for vector-ready intelligence. Configure a collection and (optionally) schema; workflow or API writes smart bites and embeddings to that collection. Supports source lineage and metadata for Vector Search and grounded answers. See Vector Catalog overview and Collections.
Object storage
| Connector | Use for |
|---|---|
| Amazon S3 | Write JSON or parquet output for analytics, backup, or custom indexing. See AWS. |
| Azure Blob Storage | Same as S3; use for Azure-centric pipelines. See Azure. |
| Google Cloud Storage | Same as S3; use for GCP. See GCP. |
Use when you want to keep raw extraction output in your own bucket for compliance, re-processing, or custom vector stores.
Your vector store
- Where supported, destination connectors can write to your vector database (e.g. via a dedicated connector or webhook). Check Reference: Connectors for available destinations and parameters.
- Alternatively, write to object storage and use your own job to load into your vector DB; preserve source lineage in metadata.
Configuration
- Credentials — Same as source connectors: API key, OAuth, or IAM. Least-privilege write access. See Authentication.
- Path or collection — Specify the bucket/prefix or collection name so output lands in the right place.
- Format — JSON, parquet, or collection-specific format. Ensure schema and source lineage fields are included when needed for downstream use.
Workflows
- In a workflow, the final step (or a dedicated “deliver” step) sends extraction output to one or more destinations. See Workflows overview.
- Monitor run history so failed deliveries are retried or fixed. See Monitoring.
Common pitfalls
- Wrong collection or path — Double-check destination config so smart bites don’t land in the wrong collection or bucket.
- Missing lineage — Ensure document ID and run ID (or equivalent) are written so grounded answers and audit remain traceable. See Vector Catalog: Lineage.
- Credential expiry — Rotate keys and update connector config; monitor for delivery failures.
Next steps
- Source connectors — Ingest from storage and apps.
- Reference: Connectors — Full list and parameters.
- Vector Catalog: Collections — Configure catalog destinations.