Any support with voyage 3 context

[superbaim]

Hi,

is there a support with the latest voyage 3 context embedding ? I dont see the model in prod version.

Thanks

[dosubot[bot]]

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

Yes, RAGFlow supports Voyage 3 context embedding as of version 0.20.0, released August 2025. The production configuration includes Voyage 3 and its variants (like voyage-3-large, voyage-3.5, voyage-multimodal-3) as embedding models under the "Voyage AI" provider. You can specify these in your config files, for example: embedding_model: "voyage-multimodal-3@Voyage AI" as shown in the docs and config examples. Make sure your deployment is running v0.20.0 or later and that your llm_factories.json includes the Voyage 3 models—older versions may not show them or could have bugs with embedding shape mismatches [release notes][config example][embedding implementation]. If you still don’t see the model, double-check your configs and version, and let me know if you need help troubleshooting.

To reply, just mention @dosu.

---

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

[superbaim]

are you sure, voyage context embedding need all_chunk(chunk all merged) and instruction for document context

[dosubot[bot]]

No, RAGFlow does not require merging all chunks into one (all_chunk) or adding a special instruction for document context when using Voyage 3 embeddings. The pipeline splits documents into chunks and embeds each chunk individually using the VoyageEmbed class, which calls the Voyage API with input_type="document" for each chunk—there’s no logic to merge all chunks or inject extra instructions for context in the embedding step. This approach is consistent across embedding models in RAGFlow and is confirmed by the code implementation [source]. If you have a specific use case that needs merged context or custom instructions, it would require custom preprocessing outside the default pipeline.

To reply, just mention @dosu.

---

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