Overview
Vellumiq is a production-grade Retrieval-Augmented Generation (RAG) pipeline designed to address common failure modes of naive LLM wrappers, such as temporal hallucinations, high token costs on repeat queries, and vocabulary mismatch. It is built with FastAPI, Qdrant, FastEmbed, and Claude 3.5. The system isolates retrieval and generation logic to enable independent horizontal scaling of compute and database resources. It utilizes Claude 3.5, Qdrant, FastAPI, and Langfuse/Sentry for observability. The pipeline includes features like multi-hop query decomposition, query normalization, semantic caching, temporal hybrid search, cross-encoder reranking, and corrective RAG with pruning. It also provides API resilience and cost control mechanisms, such as rate limiting and client disconnect handling. The frontend renders citations mapping to deterministic chunk IDs, and the repository includes evaluation and CI/CD pipelines with GitHub Actions.
Pricing
Free
Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.
Visit WebsitePrices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.
Use Cases
- Teams use Vellumiq to handle complex multi-part questions by automatically decomposing them into sub-queries and fusing results.
- Teams use Vellumiq to reduce token costs on repeat queries through semantic caching that returns answers in under 100ms at zero LLM cost.
- Teams use Vellumiq to enforce point-in-time accuracy by applying temporal metadata constraints during hybrid search.
- Teams use Vellumiq to improve retrieval relevance by re-scoring candidate chunks with a local cross-encoder model.
- Teams use Vellumiq to prevent unbounded contextual indexing costs by rate-limiting ingestion endpoints.
- Teams use Vellumiq to ensure zero hallucinated sources by rendering citations that map to deterministic chunk IDs.
- Teams use Vellumiq to catch regressions in RAG quality through continuous integration pipelines that evaluate multiple metrics.
Features
Retrieval-augmented generation pipeline
Horizontally scalable architecture
Temporal hybrid search
Semantic response caching
Dynamic query condensation
Reviews
No reviews yet. Be the first to share your experience!