Converse

Enables users to upload or link documents (PDFs, Word docs, web pages) and ask natural language questions about their content through a chat interface. The system parses document content into embeddings, stores them in a vector database, and uses retrieval-augmented generation (RAG) to ground LLM responses in the source material, ensuring answers cite specific sections rather than hallucinating.

Generates LLM responses that are explicitly grounded in retrieved document passages, with automatic citation of source locations (page numbers, section headers). Uses a citation-aware prompt template that instructs the model to reference specific excerpts, reducing hallucination and enabling users to verify answers by jumping to source material.

Allows users to upload multiple documents and ask questions that synthesize information across all of them using semantic similarity search. The system embeds all documents into a shared vector space, retrieves relevant passages from multiple sources for a single query, and generates unified responses that integrate information across documents while tracking which document each fact came from.

Allows users to paste URLs or web links directly into Converse, which automatically fetches, parses, and indexes web page content for querying. The system extracts text from HTML, removes boilerplate (navigation, ads, footers), and treats web content identically to uploaded documents, enabling conversation with live web pages without manual copy-paste.

Generates summaries of uploaded documents at user-specified granularity (brief one-liner, paragraph summary, detailed outline). Uses prompt-based summarization where the LLM is instructed to extract key points at the requested detail level, optionally constrained by token limits to ensure concise output. Summaries are generated from the full document context rather than just retrieved passages.

Maintains conversation history within a document session, allowing users to ask follow-up questions that reference previous answers without re-stating context. The system retains the conversation thread, previous retrieved passages, and user intent across multiple turns, enabling natural multi-turn dialogue about document content.

Allows users to maintain separate conversation threads for different documents, with automatic context isolation to prevent information leakage between documents. When switching documents, the system clears the previous document's context and starts a fresh conversation, preventing the LLM from conflating information across unrelated documents.

Offers a free tier with limited document uploads, query quota, and document size limits, with paid tiers unlocking higher limits and premium features. The system tracks usage metrics (documents uploaded, queries executed, storage used) and enforces soft limits that encourage tier upgrades without completely blocking free users.