Back to the changelog

• Feature

  Tenant document library

  Standard contracts, manuals, reference materials: a dedicated page where you upload them once and call them anywhere with a stable slug — `{{library.standard-contract.link}}` in a webhook, an AI prompt or an event. Versions, trash with restore, built-in viewer preview, and per-role write permissions.

• Feature

  Text extraction for AI

  FlowSharp reads PDF, Word and text files and keeps their text for your flows: enable it with a flag on the document field, the `extractText` script function, or by marking a library asset. Then use `{{field.text}}` in AI prompts and templates — with an always-visible status badge. Never a surprise extraction: it only runs when you ask.

• Integration

  Text and library for automations too

  n8n and external agents can now upload library assets from a URL, list them and read extracted text with a single call (`GET /documents/:id/text`): instant when cached, on-the-fly for small files. New `full` key scope for tenant-level operations and two new MCP tools: the catalog grows to 29.

• Feature

  Flow Agent knows your documents

  Mark a library asset as "agent context" and the assistant sees it in its horizon: it reads the text page by page, only when the conversation needs it — answers grounded in your materials, without bloating every exchange.

• Security

  Hardened extraction and a tougher document channel

  File parsing runs in an isolated environment with real memory limits and timeouts, and "bomb" archives are rejected before even being opened. Ingest source URLs travel encrypted only, permanent origin errors are no longer retried pointlessly, and PDF preview in the viewer is back without loosening the anti-embedding defenses.

• Design

  Always on the right flow version

  "Back to flow" from an item's detail now takes you to the current version; and if you open a superseded version, a banner tells you and takes you to the right one in one click.