[Question]: Clarification on metadata_condition in RAGFlow: Does it filter by the original document's update/creation time?

[Chinaniu]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

Body:

Following up on the implementation of time-based filtering, I need to clarify the source and stage where the metadata_condition filter is applied.

My understanding and question are based on this code:

metadata_conditions = {
"logic": "and",
"conditions": [
{
"name": "update_time", # Which update_time?
"comparison_operator": ">=",
"value": str(seven_days_ago)
}
]
}

Core Question 1: What does "update_time" refer to?

When I filter using {"name": "update_time"}, am I filtering based on:

A) The original document's filesystem metadata (e.g., the mtime or ctime from when the file was last modified or created on the source system)?

B) The timestamp from when the document was uploaded or processed into the RAGFlow knowledge base?

C) A custom metadata field that needs to be manually extracted and assigned during document ingestion?

Core Question 2: At which stage does this filter take effect?

Does the metadata_condition directly instruct the vector retrieval engine to only consider chunks from documents meeting the time criteria? Or is it a post-retrieval filter applied after the similarity search?

Why this matters for my use case:
We are implementing a "recent documents only" feature. It’s crucial that the filter works on the document's actual business relevance date (e.g., a report's publication date or last edit date), not just the system processing time. Clarity on this point will determine if we need to pre-process our documents to inject a custom timestamp field.

Could you please point me to the relevant documentation or explain the expected behavior? Thank you!

[dosubot]

The update_time field in RAGFlow refers to the timestamp when the document was last updated within the RAGFlow system—such as during ingestion, processing, or when metadata is changed—not the original file's creation or modification time on disk. If you want to filter by a business-specific date (like publication date), you need to extract that value and inject it as a custom metadata field (e.g., in meta_fields) during ingestion or via the metadata update API.

The metadata_condition filter is applied during the retrieval stage: it limits which documents or chunks are considered for vector or keyword matching, so only those meeting your filter criteria are retrieved—this is not a post-retrieval filter. Filtering only works if the field you want to filter on exists in every document's metadata and matches the filter name exactly. Passing explicit document_ids disables metadata filtering. For more details, see the HTTP API reference and Document model code.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[Chinaniu]

@dosu Clarification on update_time in metadata_condition: Does it filter only on system time, or also on dates within document content?

Thank you for the previous clarification. I need to confirm the exact scope of the metadata_condition filter.

When I use a filter like the one below, what exactly is being compared against the value?
{
"name": "update_time",
"comparison_operator": ">=",
"value": "1704067200"
}

My specific question is: Does this filter only evaluate against the document's system update_time metadata (the timestamp RAGFlow stores when the document is processed), or does it also somehow consider dates found within the document's textual content itself (e.g., a "2023 Annual Report" mentioning dates like "Q1 2023" or "last updated on 2023-10-01")?

[dosubot]

The metadata_condition filter with name "update_time" only compares against the system-stored update_time field—the timestamp when the document was last updated or processed in RAGFlow. It does not look at or extract dates from the document's textual content (like "Q1 2023" or "last updated on 2023-10-01") unless you explicitly extract those dates and store them as custom metadata fields during ingestion or via the metadata update API. If you want to filter by a business-specific date found in the document content, you need to preprocess your documents to add that date as a custom metadata field (for example, "publication_date") and then filter on that field instead of "update_time" reference docs.

Let me know if you want tips on how to extract and inject custom dates for filtering!

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[Magicbook1108]

dosu is correct.