Solutions
·RAG
Build RAG pipelines
that stay fast and private.
Embed with text-embedding-3 or BGE via API. Store vectors in Qdrant, Weaviate, or Chroma on dedicated instances. Generate answers with Claude 4, DeepSeek-V3.2, or any of 200+ models. Zero egress fees -- your data never leaves the platform.
The RAG Stack
Embed, store, retrieve, generate. All here.
Embedding models via API
text-embedding-3-large, BGE-M3, E5-Mistral, and more via a single endpoint. Generate dense or sparse embeddings for text, code, and images.
Vector databases on instances
Run Qdrant, Weaviate, Chroma, or Milvus on dedicated instances with persistent storage. Full root access for custom configuration.
Generation models via API
Claude 4 Sonnet, DeepSeek-V3.2, Llama 4 Scout, Gemini 2.5 Flash -- use any model for the generation step. Switch models without changing your pipeline.
Zero egress, zero latency tax
Embeddings, vector DB, and generation all run on the same platform. No cross-provider network hops, no data transfer fees eating your margins.
Chunking and preprocessing
Use dedicated instances to run custom chunking pipelines -- LangChain, LlamaIndex, or your own scripts. Full Python/Node environment with Docker support.
Scale each layer independently
More embedding throughput? Call the API. More vector storage? Expand your instance disk. More generation capacity? Add models. No over-provisioning.
Models & Tools
Popular choices
for RAG pipelines.
Mix and match embedding models, vector stores, and generation models to fit your use case.
text-embedding-3-largeEmbedding · APIHigh-accuracy dense embeddings
BGE-M3 / E5-MistralEmbedding · APIMultilingual, hybrid retrieval
Qdrant / Weaviate / ChromaVector DB · InstancePersistent vector storage
Claude 4 / DeepSeek-V3.2Generation · APIContext synthesis and answers
Llama 4 Scout / Qwen3Generation · API or self-hostedOpen-source generationHow It Works
Three steps to a RAG pipeline.
Embed your documents
Call the Inference API with your text to generate embeddings. Use text-embedding-3, BGE, E5, or any supported embedding model. Per-token pricing, no minimums.
Store in a vector database
Deploy Qdrant, Weaviate, or Chroma on a dedicated instance. Index your embeddings with persistent storage. Zero egress means retrieval stays fast and free.
Retrieve and generate
Query your vector DB, pass context to Claude 4, DeepSeek-V3.2, or any generation model via the same API. Entire pipeline runs on-platform with no data transfer fees.