Anatole KABORE

Agentic Knowledge Workspace turns an organisation's scattered knowledge into a source you can question. It accepts text, PDFs, scanned pages read by OCR, and audio transcribed by the model: a spoken sentence becomes an indexed document in about nine seconds. Answers are grounded in hybrid retrieval — semantic plus lexical — and always cite the document and the passage they come from. When the workspace holds nothing relevant, the system refuses instead of guessing: a measured relevance threshold makes it name the workspace and ask for the missing document. Beyond answering, an agentic task plans its research, scopes it to the selected documents, drafts a note, verifies every claim against its source, corrects itself, and checks that the result answers the instruction. The result exports to DOCX or PDF. Governance is built in: two-level roles, workspaces that scope what the assistant can see, per-document access control on every route that takes a document id, SHA-256 versioning, and an audit journal. 124 automated tests. PHP 8.4.27, MySQL 8.4.13, no framework.