Chat Citation Issues

[Yeon1007]

There's a problem with the "show citation" feature in chat: For files parsed and chunked using a pre-defined chunking method (built-in/pipeline) set in the dataset, when displaying citations in chat answers, clicking on a file within the citation highlights the cited portion in yellow. However, manually added chunks can only be cited and not highlighted in the original text. Therefore, the following questions arise:

1. Can manually added chunks(newly added) also be highlighted in yellow in the original text when cited?
2. For chunks parsed using a pre-defined chunking method, if these chunks are subsequently modified, can they be highlighted in the original text when cited (not as newly added chunks, but just modifications to existing chunks)? @dosu

[dosubot[bot]]

Manually added chunks can't be highlighted in the original text when cited because they lack the positional metadata (the positions field) that maps them to specific locations in the source document. Only chunks created by the built-in/pipeline parser include this metadata, which is what enables the highlight feature in the chat UI. Without these coordinates, the system can't visually highlight the cited region in the original file—manual chunks can be cited, but not highlighted in context (source).

If you modify a pre-defined (pipeline) chunk, the ability to highlight depends on whether the positional metadata is preserved and still matches the new content. The current API for modifying chunks updates the content and tokenization but does not update or guarantee the accuracy of the positions field. If the positions metadata is missing or no longer matches the chunk's content, highlighting will not work or may be inaccurate (source).

To enable highlighting for manually added or modified chunks, the backend would need to let users (or the system) supply or update the positions metadata for those chunks, and the frontend would need to consume this metadata as it does for pipeline chunks. There is community interest in supporting user-maintained metadata for custom chunks, but this isn't implemented yet (source).

To reply, just mention @dosu.

---

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

   

[dosubot[bot]]

Yes, you can use different chunking methods for different files within the same dataset in RAGFlow. When you change the chunking method in the dataset configuration, new files uploaded afterward will use the new method, while previously uploaded files keep the chunking method they were originally processed with. The chunking method is stored per file, not just at the dataset level, so each document can have its own chunking configuration. Files uploaded in the same batch use the same method, but if you change the configuration between uploads, subsequent files will use the updated method. The "general" chunking method also automatically selects the best parser for each file type, so mixed strategies are possible even within a single upload details discussion thread.

You can't retroactively change the chunking method for files already uploaded—if you want a different method for those, you'll need to re-upload or re-chunk them.

To reply, just mention @dosu.

---

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

   

[Yeon1007]

how to connect Langfuse to RAGflow, and I have the secret key and public key, how should I fill in the host URL? I'm self hosting RAGflow and Langfuse locally. @dosu

[dosubot[bot]]

To connect your self-hosted Langfuse to RAGflow, go to the RAGflow UI, open "User Settings → System" (or sometimes "User Settings → API" at the bottom), and find the Langfuse configuration card. Enter your Langfuse public key, secret key, and for the host URL, use the base URL of your Langfuse instance, including protocol and port if needed—for example, http://localhost:3000 or http://127.0.0.1:3000 if Langfuse is running on port 3000 locally. After saving, RAGflow will start sending traces to Langfuse automatically—no code changes required. Make sure RAGflow can reach Langfuse on that address (check Docker/network settings if needed) [docs].

If you see an authentication error, double-check your keys and host URL.

To reply, just mention @dosu.

---

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

   

[Yeon1007]

it show ConnectError('[Errno 111] Connection refused') @dosu

[dosubot[bot]]

A ConnectError('[Errno 111] Connection refused') means RAGflow can't reach your Langfuse server at the host and port you configured. The most common reasons are: Langfuse isn't running, it's running on a different port, the host URL in RAGflow doesn't match, or there's a network restriction (like Docker networking or a firewall) blocking access.

Check these:

• Make sure the Langfuse server is running and listening on the port you specified (for example, http://localhost:3000).
• Double-check that the host URL in RAGflow exactly matches the address and port where Langfuse is reachable.
• If you're running either service in Docker, verify that the containers are on the same network and can talk to each other (localhost inside a container is not the same as localhost on your host machine).
• Look at the Langfuse server logs and RAGflow logs for more details about the failed connection attempt.

If you fix the network or port issue, RAGflow should be able to connect—no code changes are needed, just correct configuration and network access. The host URL must include the protocol (http/https) and port if it's not the default reference.

To reply, just mention @dosu.

---

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

   

[Magicbook1108]

Yeah dosu is correct,

1. Manually added chunks can't be highlighted in the original text when cited
2. You can have a try.