MoshiRAG
MoshiRAG adapts Moshika for asynchronous knowledge retrieval. It can request information and receive text while the speech conversation continues, rather than requiring a complete stop for each search.
Research release
Retrieval requires additional components
The documented system uses the frozen ARC4 encoder and a separate STT model. A download of the front end does not supply an indexed knowledge base or guarantee that retrieved information is correct. The artifact repository contains shared components, not another dialogue model.moshika-rag-pytorch-bf16: Kyutai model card
The ARC4 dependency retains the applicable Llama terms. This does not change the separate speech front-end license.ARC4_Encoder_Llama: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/moshika-rag-pytorch-bf16kyutai/moshika-rag-candle-bf16 |
| Release scope | Speech front-end model; the retrieval back end is separate.moshika-rag-pytorch-bf16: Kyutai model card |
| License | CC BY 4.0.moshika-rag-pytorch-bf16: Kyutai model card |
Related models
- ARC4 Encoder Llama
Kyutai · ARC Encoder
A text encoder with fourfold pooling trained for use with a Llama decoder.
Launch date not confirmedResearch releaseDownloadable - Kyutai STT 1B English/French
Kyutai · Kyutai STT
A streaming English and French transcription model with a configured output delay of 0.5 seconds.
Launch date not confirmedResearch releaseDownloadable - Moshi Moshika
Kyutai · Moshi
A Moshi speech-dialogue model that listens while speaking, using Kyutai's Moshika voice.
Launch date not confirmedResearch releaseDownloadable