Working with metadata

[bodicka2002-wq]

Where did the interactive metadata changes go? I'm creating files in a dataset. I am adding metadata to these files. Previously, it was possible to use metadata to make a selection so that the chat could search for information. only for certain files, certain metadata. But now, interactively, in the chat, in the settings, I can't find it. Has it been moved somewhere? How to work with metadata correctly then now? Or what is the right way to make a selection now? I added metadata to the files, how do I select them now?

[dosubot[bot]]

Hi @bodicka2002-wq! I'm Dosu and I’m helping the ragflow team.

The metadata filtering feature in chat settings has not been removed — it was broken by a UI bug caused by an internal refactoring that renamed kb_ids to dataset_ids. Some UI components weren't updated to watch the new field name, which caused the metadata filter options to disappear from the chat settings page [1]. This was fixed in April 2026, so upgrading to the latest version should bring back the metadata filtering UI in chat settings.

How metadata filtering works in current versions:

1. Define metadata at the dataset level — add metadata fields with name, type, and description.
2. Attach metadata to your documents via meta_fields (through the UI or API).
3. In chat/assistant settings, once you select your datasets, a metadata filter section should appear with these modes [2]:
   • Disabled — no filtering
   • Automatic — the LLM auto-generates filter conditions from the user's question and available metadata [3]
   • Semi-Automatic — you choose which metadata keys the LLM considers [4]
   • Manual — you explicitly define filter conditions (key-operator-value with AND/OR logic) [5]

Important notes:

• Make sure field names match exactly between your document metadata and filter definitions [6].
• If you added metadata to already-parsed documents, you may need to re-parse with the "Apply Auto Metadata Settings" option checked [7].
• The metadata filter UI only shows when datasets are selected in the chat configuration [8].

What version of RAGFlow are you currently running? If you're on an older version (pre-April 2026), upgrading should resolve the missing metadata UI.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[musaabhasan]

Even when the UI changes, I would think of metadata as part of the retrieval contract rather than a display-only document property.

The workflow I would use is:

1. Define stable metadata keys first, for example brand, document_type, language, version, region, effective_date.
2. Add metadata before or during ingestion so chunks inherit it consistently.
3. Verify that metadata exists at chunk level, not only at file level.
4. Use metadata filters in the assistant/dataset retrieval configuration where available.
5. Test a query with and without the filter and compare retrieved chunk ids.

If the interactive selector is temporarily broken or moved, the key thing to confirm is whether metadata is still present in the indexed chunks. If it is, the issue is UI/configuration access. If it is not, the document may need to be reprocessed so chunk-level metadata is attached. For reliable production use, I would also keep a small metadata test set: one query that should return only a specific brand/version/language and one query that should return nothing when the filter is applied.

[Ganesh-403]

@bodicka2002-wq From what I can tell, the metadata filtering functionality hasn't been removed—it has just moved into the assistant's retrieval configuration in newer versions.

The current workflow is:

1. Define metadata fields for your dataset (name, type, description).
2. Assign metadata to your uploaded files.
3. Configure metadata filtering in your Chat Assistant after selecting the dataset(s). You can choose from:
   • Disabled
   • Automatic (LLM generates filters from the user's query)
   • Semi-Automatic (LLM only considers selected metadata fields)
   • Manual (explicit filter conditions such as brand = Canon) :contentReference[oaicite:0]{index=0}

If you're not seeing those options, a few things are worth checking:

• Which RAGFlow version are you running?
• Have you already selected the dataset(s) for the assistant? The metadata filtering options only appear after datasets are associated with the assistant. :contentReference[oaicite:1]{index=1}
• Were the metadata fields added before or after the documents were parsed? If metadata was added after ingestion, you may need to reprocess or reparse the documents so the indexed chunks inherit the metadata correctly. :contentReference[oaicite:2]{index=2}

If your goal is to search only a subset of files (for example, brand = HP or language = de), metadata filtering is still the recommended approach. The official documentation describes configuring metadata at the dataset level and then using retrieval-time metadata filters in the assistant. :contentReference[oaicite:3]{index=3}

Could you also share:

• your RAGFlow version,
• and whether you're referring to the Chat Assistant UI or the Retrieval Test page?

That would help determine whether you're encountering a UI change, a configuration issue, or a version-specific bug.

If this solves your problem, feel free to mark it as the accepted answer so others can find it easily.