Agentset.ai

Agentset is an open source platform for building AI chat and search on a company's own documents. It handles the retrieval augmented generation pipeline end to end, from extraction and chunking through retrieval and agentic search, so a developer can ship a grounded question answering experience without first becoming a retrieval specialist. Answers carry citations automatically, and a reader can open the retrieved chunk behind any claim through an inline source pill. It is aimed at engineering teams putting a knowledge base in front of users. JavaScript and Python SDKs ingest more than twenty two file formats, covering PDF, DOCX, XLSX, PPTX, HTML, Markdown, CSV, plain text, email formats and images, and the system treats images, graphs and tables as answerable content rather than discarding them. Metadata filtering narrows answers to a subset of the corpus. The platform is model agnostic: the vector database, embedding model and language model are all chosen by the user, with support for providers including OpenAI, Anthropic, Google, Mistral, Cohere, Pinecone and Qdrant. An MCP server exposes the knowledge base to external applications, there is an AI SDK integration, and every namespace gets a prebuilt hosted chat interface with a custom domain, email or domain allowlists and configurable models. The Free plan costs $0 and covers one thousand pages and ten thousand retrievals with community support, but no connectors and no custom integrations. Pro is $49 per month for ten thousand pages and unlimited retrievals, with additional pages at $0.01 each, connectors at $100 each per month, deep research and email support. Enterprise is quoted and adds unlimited pages and retrievals, on premise or bring your own cloud deployment, custom formats, single sign on, SOC 2, HIPAA and GDPR reports and dedicated engineering support. A page counts as every one thousand characters of parsed text.