Source sync and indexing

Use this page when the question is not what type of source should I create?, but why is an existing source not synchronized or searchable yet? AI Assistant depends on two separate steps for much of its knowledge: obtaining the source content, then making that content available for retrieval.

​

Zendesk synchronization

Zendesk-backed resources are synchronized from the connected Zendesk account. This includes resources such as:

• Help Center Articles;
• Ticket Fields; and
• Macros.

Open AI Assistant → Zendesk → Setup & Sync to review the connection and synchronization state. A resource may be absent because the Zendesk credential cannot access it, the source is not available to that account, or synchronization has not completed since the resource changed.

​

Public Help Center fallback

When RAUM cannot reach Zendesk’s staff Help Center API with the connected credential, it can fall back to the public Help Center feed. That fallback can only see published, unrestricted content available publicly. It cannot manufacture access to private Guide content the credential does not have permission to read.

​

Processing uploaded knowledge

Uploaded documents and crawled pages move through processing states before they are ready for normal retrieval. Common states include:

• Processing;
• Ready; and
• Error.

Use reprocessing when a document failed or needs its processed representation rebuilt.

​

Embeddings and semantic retrieval

Searchable resources rely on embeddings so RAUM can retrieve by meaning rather than exact keyword match. A source can be present and active but still fail to participate normally if its embedding is missing. The Assistant Sources workspace includes repair tooling for procedures, Help Center articles, and AI Guidelines with missing vectors. When repair fails, RAUM reports the provider error rather than treating the content as successfully indexed.

​

After changing the embedding model

Changing the embedding provider/model can affect retrieval consistency. After such a change:

1. verify the new embedding model is active;
2. run the available embedding repair/backfill workflow;
3. reprocess failed document sources as necessary;
4. test representative tickets; and
5. inspect Ticket Activity to confirm the expected resources are retrieved.

​

Diagnose a missing source

When Assistant appears to ignore content, check in this order:

1. Does the source exist?
2. Is it active?
3. Did synchronization/processing succeed?
4. Does it have the required embedding/index state?
5. Does targeting exclude the current ticket?
6. Is a more relevant/conflicting source being selected instead?
7. What does Ticket Activity say was actually used?

This sequence is more useful than repeatedly rewriting the ticket until retrieval happens by coincidence.