Agentset

About Agentset

Agentset is an open-source platform for building AI chat and search over your own data without assembling a RAG stack yourself. It covers the whole path — ingestion and chunking across 22-plus file formats, embedding, retrieval, agentic search and answer generation — behind JavaScript and Python SDKs, so a team that needs a reliable answer engine on top of a document corpus can ship one without hiring for retrieval expertise. Three design choices distinguish it from a hand-rolled pipeline. Multimodal handling means images, graphs and tables inside documents are treated as first-class retrievable content rather than being dropped at parse time, which is where most naive pipelines quietly lose half a knowledge base. Citations are automatic, so every answer carries inspectable sources — the single most effective mitigation for hallucinated answers in a customer-facing deployment. And metadata filtering lets a query be scoped to a subset of the corpus, which is what makes per-tenant or per-permission answering possible. The project publishes benchmark positions on MultiHopQA and FinanceBench and provides customisable preview links so non-technical stakeholders can test a deployment and leave feedback without an account. Supported inputs include PDF, DOCX, XLSX, PPTX, HTML, CSV, Markdown, email formats and common image types.

Key Features

Tags