Can I use RAGFlow together with Langchain?

[kenmbelenzi]

How can I integrate RAGFlow as an external retriever for a LangChain-based application? Specifically, is there an official or recommended API/SDK pattern for using RAGFlow purely as a context provider (vector retrieval layer) while LangChain handles orchestration and LLM calls?”

[dosubot[bot]]

Hi @kenmbelenzi! I'm Dosu and I’m helping the ragflow team.

Yes, you can use RAGFlow with LangChain by integrating RAGFlow as an external retriever/context provider. The recommended approach is to use RAGFlow’s REST API or Python SDK to fetch relevant context chunks from your knowledge base, then pass those results into LangChain for orchestration and LLM calls. There’s no dedicated LangChain integration, but the API/SDK pattern is straightforward: call RAGFlow’s /retrieval endpoint (or use the retrieve method in the Python SDK) with your query, API token, and knowledge base ID, and use the returned chunks as context in your LangChain pipeline. This pattern aligns with how LangChain expects external retrievers to work, and RAGFlow’s API is designed for this use case (SDK example, API details, FAQ). All API usage is self-hosted, so you’ll need to deploy your own RAGFlow instance. There’s no nonstandard integration pattern—just use standard HTTP or Python client calls to retrieve context.

To reply, just mention @dosu.

---

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

[dosubot[bot]]

RAGFlow’s Ollama integration is model-agnostic—it will use any embedding model your Ollama server exposes, as long as you specify the model name and provider in the format <model_name>@Ollama. However, not all models in Ollama’s library support embedding via the API. The only embedding model explicitly recommended and tested in the RAGFlow docs is bge-m3 (567M), which you can pull and use as bge-m3@Ollama (docs).

If a model like nomic-embed-text doesn’t work, it’s likely because Ollama doesn’t expose it as an embedding endpoint, not because of a restriction in RAGFlow (embedding_model.py). To see which embedding models are available, run ollama list and check the Ollama model library for embedding-capable models. For best results, use bge-m3@Ollama for embeddings in RAGFlow.

To reply, just mention @dosu.

---

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

[kenmbelenzi]

@dosu I am getting error
hint : 102
Fail to access model(Ollama/bge-m3:latest).ERROR: CONNECTION_ERROR - litellm.APIConnectionError: Ollama_chatException - [Errno 111] Connection refused

when setting embedding I am getting error
hint : 102
Fail to access embedding model(bge-m3:latest).Failed to connect to Ollama. Please check that Ollama is downloaded, running and accessible. https://ollama.com/download

[dosubot[bot]]

The "connection refused" error means RAGFlow can't reach your Ollama server at the configured address. Here are the most common causes and fixes:

• Ollama must be running and listening on the correct host/port (default is http://localhost:11434). If you're running RAGFlow and Ollama in Docker or on different machines, use the correct network address (like http://host.docker.internal:11434 for Docker on Windows/Mac, or your host's LAN IP if remote) and make sure port 11434 is open and accessible. If Ollama is only listening on 127.0.0.1, set OLLAMA_HOST=0.0.0.0:11434 and restart Ollama to accept external connections.
• The embedding model (bge-m3:latest) must be pulled and available in Ollama (ollama pull bge-m3).
• Your RAGFlow embedding model config must use the correct format: bge-m3:latest@Ollama and the base_url must match your running Ollama instance, with no trailing slash.
• If you're on Linux or using Docker, check that both containers are on the same network and can communicate. You can test connectivity from the RAGFlow environment with curl http://host.docker.internal:11434/ or the appropriate IP.
• Ollama only runs on macOS 14+; if you're on Windows or Linux, you can't run Ollama locally and will always get connection errors unless you connect to a remote macOS host running Ollama.

If you're using RAGFlow v0.20.4, there is a known bug causing Ollama connection errors even with correct config. This is fixed in the nightly build and v0.20.5+—upgrading resolves the issue. If you can't upgrade, make sure your base_url has no trailing slash, or downgrade to v0.20.3, which is confirmed to work with Ollama models. See detailed troubleshooting and version notes in these threads: issue 10050, issue 9806, issue 7542, issue 9829.

If you share your OS, deployment method (Docker, local, remote), and config details, I can help you pinpoint the exact fix.

To reply, just mention @dosu.

---

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

[kenmbelenzi]

@dosu I am having a challenge parsing my pdf document
19:30:39 Task has been received.
19:30:39 Page(18): Start to parse.
19:30:39 Page(18): OCR started
19:30:42 Page(18): Processing...
19:30:42 Page(18): OCR finished (2.51s)
19:31:08 Page(18): Layout analysis (26.03s)
19:31:08 Page(18): Table analysis (0.46s)
19:31:08 Page(18): Text merged (0.00s)
19:31:09 Page(18): Finish parsing.
19:31:09 Page(18): Generate 7 chunks
19:31:10 Page(18): [ERROR]Generate embedding error:do embedding request: Post "http://127.0.0.1:58047/embedding": EOF (status code: 500)
19:31:10 [ERROR][Exception]: do embedding request: Post "http://127.0.0.1:58047/embedding": EOF (status code: 500)

[dosubot[bot]]

The EOF (status code 500) error during embedding usually means the embedding service crashed or closed the connection unexpectedly—often due to a timeout, resource limits, or misconfiguration. This is a common issue in RAGFlow when the embedding step takes too long or the service isn't running as expected.

Here’s what you can do:

• Make sure your embedding service is running and accessible at the configured address (in your case, http://127.0.0.1:58047/embedding). Try opening that URL or using curl to check connectivity.
• Check the logs of your embedding service for crashes or errors.
• Lower the EMBEDDING_BATCH_SIZE in your .env file (try 2 or 4), and set MAX_CONCURRENT_TASKS=1 to reduce load. Disabling non-essential features (like reranker, RAPTOR, auto-keyword, auto-question) can also help.
• If you’re comfortable editing code, you can increase the embedding timeout by changing the @timeout(5) decorator on the batch_encode function in rag/svr/task_executor.py to a higher value (like 15 seconds), then redeploy RAGFlow. This is needed if your environment or model is slow to respond [source].
• If you’re running in Docker or WSL2, make sure you have enough CPU/memory allocated, and restart your containers after making changes [source].
• If embedding is still unreliable, consider using an external embedding service (like vLLM, Ollama, or Xinference) [source].

If you share your embedding model/provider config and deployment setup (local, Docker, OS), I can help you pinpoint the exact fix.

To reply, just mention @dosu.

---

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

[kenmbelenzi]

[dosubot[bot]]

The error means your Docker container can't find the rag package because it's not in the Python path inside the container. This usually happens if the rag directory isn't copied into the image, your WORKDIR isn't set to the project root, or PYTHONPATH isn't set correctly.

To fix this, make sure your Dockerfile does the following (following RAGFlow's official pattern):

• COPY the rag directory (and any other needed source folders) into the container, e.g.:

  COPY rag rag
  

• Set the working directory to the project root (where rag lives):

  WORKDIR /ragflow
  

• Set the PYTHONPATH to the project root:

  ENV PYTHONPATH=/ragflow
  

• Start your server from the project root (e.g., python main.py or via an entrypoint script).

If your custom recommendation folder is outside the project root or you run code from a subdirectory, Python won't find rag unless you adjust PYTHONPATH to include the parent directory of rag. You can also add this at the top of your main.py for a quick test:

import sys, os
sys.path.insert(0, os.path.abspath(os.path.dirname(__file__)))

But the best practice is to match the official Docker setup: project root as WORKDIR, PYTHONPATH set, and all source folders copied in. This will resolve the import error for rag and any other internal modules.
Reference: official Dockerfile and launch scripts

To reply, just mention @dosu.

---

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

[kenmbelenzi]

@dosu it doesnt work

1 | FROM python:3.11-slim

2 | >>> COPY rag rag

3 | WORKDIR /ragflow

4 |

---

failed to solve: failed to compute cache key: failed to calculate checksum of ref da58j6w10277jrplf39aioj0g::nczegjqtxlyi3hhebft2adpac: "/rag": not found

here is my docker file

FROM python:3.11-slim
COPY rag rag
WORKDIR /ragflow

Install PostgreSQL client libraries required by psycopg2-binary

RUN apt-get update && apt-get install -y
libpq-dev
gcc
&& rm -rf /var/lib/apt/lists/*
ENV PYTHONPATH=/ragflow

RUN pip install -r requirements.txt
CMD ["uvicorn", "main:app", "--host", "0.0.0.0","--port", "8001"]

[dosubot[bot]]

The error means Docker can't find the rag directory in your build context. The COPY rag rag line only works if rag exists in the same directory as your Dockerfile (or in the directory you run docker build from).

To fix this, make sure your project folder contains both the Dockerfile and the rag directory at the top level, then run docker build from that folder. If rag is elsewhere, either move it next to the Dockerfile or update the COPY path to match its actual location.

Also, ensure any other required files (like main.py, requirements.txt, etc.) are present in the build context and copied into the image as needed. If you’re still stuck, share a tree view of your project directory so I can give more specific advice.

References: example of this error and solution, official Dockerfile structure.

To reply, just mention @dosu.

---

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

[kenmbelenzi]

give me an example i have same structure as code in this repo