Connecting a support knowledge base is only the first step. The operational test is what happens when an article changes, disappears, or should no longer be visible to a particular user.
Amazon Bedrock Managed Knowledge Base added native Salesforce and Zendesk data-source connectors on September 23, 2026. AWS says the Salesforce connector crawls knowledge articles, while the Zendesk connector covers help-center articles and community posts. The managed connectors handle crawling, metadata extraction, and incremental synchronization, replacing a custom ingestion pipeline for these sources.
What the connectors actually move
The source system remains the place where staff maintains a support article or product note. Bedrock ingests supported content into a knowledge base that an assistant can retrieve from. Metadata can help the application filter or identify a result. The connector doesn’t decide whether the model’s answer is correct, complete, or appropriate for the person asking.
For a support bot, start with a small collection of approved articles, not the whole customer-service archive. Public help pages, internal troubleshooting notes and private account records have different audiences. A crawler that reaches all three does not mean the final assistant should expose all three.
Incremental sync has a clock.
AWS documents incremental syncing for added, modified, and removed documents. A data-source sync can run on demand or be scheduled daily, weekly, or monthly. Between a change in Salesforce or Zendesk and the next successful sync, the indexed copy may still reflect the old version. A customer-facing assistant therefore needs a freshness rule for high-impact material such as prices, eligibility, and security instructions.
One practical rule is to put a source timestamp and last successful sync time beside every retrieved answer in internal testing. Create three fixtures: a new article, an edited article, and a deleted article. Query all three before and after sync, then inspect the ingestion job statistics. That proves the content lifecycle, not just that the initial connection succeeds.
Deletion and access deserve separate tests.
AWS offers a deletion-protection configuration that can skip the delete phase if a sync would remove more than a configured percentage of indexed documents. That helps prevent a bad credential or source outage from appearing as a mass deletion. It also means an intentionally removed document might remain indexed until the issue is resolved. Review deletion-protection events rather than assuming every source deletion propagated immediately.
Access control is another independent boundary. Confirm how the connector, knowledge base, and application enforce document-level permissions for your chosen setup. Test two users with different source access against the same query. A relevant retrieved article can still be unauthorized if permissions don’t carry through the retrieval path.
When native sync beats a custom pipeline
Native ingestion removes crawler maintenance for the supported source types. A custom pipeline may still be needed for unusual transforms, very specific filtering, unsupported attachments, or a stronger freshness target than the available schedule provides. Compare the two approaches on three measurable points: time from source edit to searchable result, rate of permission mistakes, and number of failed ingestion jobs that require an operator.
Our structured-content accuracy report explains why organized source material affects retrieval quality. The Vercel Connect access guide covers the related question of which user approved an agent’s access to an external service.
A rollout checklist for support teams
- Confirm the connector is available in the AWS Region and Managed Knowledge Base configuration you use.
- Create a source account with the smallest practical content scope.
- Choose a sync schedule that matches the business impact of stale answers.
- Run new, edited, deleted, and permission-restricted document tests.
- Review retrieval results and generated answers with support staff.
- Monitor sync failures, deletion-protection events, and answer corrections after launch.
Primary sources
- AWS: Salesforce and Zendesk connector announcement
- AWS: Salesforce data source
- AWS: Zendesk data source
- AWS: Synchronize a managed knowledge-base source
Checked September 24, 2026. Connector scope is from AWS; the lifecycle tests and rollout checklist are MustHave.ai guidance.
