Add Model with LM Studio Got Error

[jacksonasl-07]

When adding a model using LM Studio, even after entering the correct model name and URL, it still displays an "access failed" message. However, checking the LM Studio logs, which received the request, so why can't it connect? Even verify a previously successfully added model (with the same configuration) results in the same "access failed" error. Is this a bug in the new version of RAGFlow?

[dosubot[bot]]

This issue is likely caused by a recent change to how RAGFlow verifies models. In March 2026, RAGFlow switched model verification from synchronous to streaming responses. Since LM Studio logs show requests are being received, the connection itself works—the failure is in the response validation.

Most likely cause: LM Studio may not be returning a properly formatted streaming response that RAGFlow expects within the 10-second timeout.

Quick things to check:

1. URL format: RAGFlow automatically appends /v1 to your base URL, so enter just http://localhost:1234 (not http://localhost:1234/v1)

2. Docker networking: If RAGFlow runs in Docker, use your host machine's IP instead of localhost (e.g., http://192.168.1.100:1234)

3. Test streaming manually:

   curl -X POST http://localhost:1234/v1/chat/completions \
     -H "Content-Type: application/json" \
     -d '{"model":"your-model-name","messages":[{"role":"user","content":"Hi"}],"stream":true}'

   You should see multiple data: chunks. If you get a single JSON response, that's the problem.

4. Model name: Use the exact full path shown in LM Studio (e.g., lmstudio-community/Meta-Llama-3.1-8B-Instruct-GGUF/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf)

Since this worked previously with the same configuration, this does appear to be related to the streaming verification change in the new version. Could you share which RAGFlow version you upgraded from and to?

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[dosubot[bot]]

There's no on/off toggle — you disable Langfuse by deleting its configuration:

1. Go to User Settings → API tab
2. Scroll to the bottom of the page — you'll see a Langfuse card
3. Click the "Configuration" button (⚙️) on the card
4. In the dialog, click the "Delete" button (🗑️) and confirm
5. Restart RAGFlow after deleting

Once removed, RAGFlow will no longer attempt to connect to Langfuse during document parsing, which should resolve the [Errno 111] Connection refused error.

If you don't see a Langfuse card there, then Langfuse was never configured for your tenant and the connection refused error is coming from your embedding service instead — in which case, check the embedding service fixes I mentioned earlier.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[jacksonasl-07]

If I use v0.24.0 RAGfFlow for chunking and embedding, after this, i downgraded the RAGFlow to v0.23.1 then use this older version to generate the knowledge graph. This action will bring any nagetive impact? will it affect the volume? @dosu

[dosubot[bot]]

This is high risk and likely to cause data loss. I'd strongly advise against it. Here's why:

Database schema incompatibility: v0.24.0 introduced breaking schema changes — new columns like graphrag_task_id, raptor_task_id, pipeline_id were added, and the meta_fields column was removed from the Document table [1]. v0.23.1 won't recognize these changes and will throw errors like Table ... doesn't exist or column mismatches [2].

Docker volumes will likely need to be wiped: MySQL image version differences between v0.24.0 and v0.23.1 can cause Cannot downgrade errors, and the typical fix is docker compose down -v — which deletes all your volumes including your chunks and embeddings [3].

No backward migration scripts exist: RAGFlow uses forward-only migrations, so there's no built-in way to safely revert the schema.

The good news — knowledge graph generation itself doesn't depend on the embedding format. It reads the content_with_weight field from chunks and uses LLM-based extraction. So the chunk data format is compatible. But the surrounding infrastructure (database schema, Docker volumes) is not.

If you still need to do this, make a full backup of all your Docker volumes and databases first. But honestly, the safer approach is to stay on v0.24.0 and troubleshoot whatever issue is pushing you to downgrade. What specific problem are you hitting with knowledge graph generation on v0.24.0?

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[jacksonasl-07]

What vectorDB and GraphDB are used in RAGFLow? @dosu

[jacksonasl-07]

how to add rerank model via hugging face in RAGFlow? @dosu

[Ganesh-403]

@jacksonasl-07 This looks like a real verification / compatibility issue, not a connection issue.

Since LM Studio logs show the request arriving, RAGFlow is at least reaching the server. The failure is happening in the model access check, not in basic HTTP connectivity. That matches the maintainer-style diagnosis in the discussion: the request is received, but RAGFlow is not accepting the returned response as a valid model verification result.

A few likely causes:

1. Base URL format

   • In RAGFlow, the provider URL is usually the host root, not the /v1 path.
   • So use http://localhost:1234, not http://localhost:1234/v1, because RAGFlow typically appends /v1 itself for OpenAI-compatible providers. The docs for local model setup follow this pattern for local providers.

2. Streaming mismatch

   • If the new RAGFlow version validates models using a streaming chat completion, LM Studio must return a stream response in the format RAGFlow expects.
   • If LM Studio returns a non-streaming JSON response, or a stream that RAGFlow does not parse correctly, model verification can fail even though the endpoint is reachable. That matches the bug report in the discussion.

3. Model name parsing / special characters

   • Recent RAGFlow release notes mention a bug where LM Studio model names containing @ caused parsing failures. That was fixed in a later release, so if your model name includes @, this may be the reason.

4. Version regression

   • Because the same configuration used to work and now fails, this strongly suggests a regression in the newer RAGFlow release rather than a bad LM Studio setup. The GitHub issue you referenced is consistent with that.

A practical way to confirm where the break is:

curl -X POST http://localhost:1234/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-model-name",
    "messages": [{"role":"user","content":"Hi"}],
    "stream": true
  }'