This blog has been busy these past months, but mostly with product and tech news: releases, features, how-tos. What those posts do not show is the bigger picture taking shape behind them. So before the summer break, here is a guided tour of the lab: five stops, from a shipped MCP server to an open-source CLI, from advanced business solutions to a first glimpse of agentic plans.
One thread runs through everything below. AI only becomes dependable in the enterprise when every layer of the document stack is open to it: the repository, the viewer, the orchestration. Not a chatbot bolted on top, but a platform where documents, metadata and processes are structured objects an AI can work with, under your rules. That is what it takes to move AI from the demo to the real world, and every stop on this tour is a variation on it.
If you read one thing we published this year, make it the FlowerDocs MCP Server announcement. The short version: FlowerDocs now ships an MCP server that connects an AI assistant directly to the platform's administration layer. You describe the document classes, workflows and search forms you need in natural language; the assistant builds them, guided by the product documentation and bounded by the permissions of the connected user.
It is live today, and it has already changed how we run our own projects. What it unlocks for you is simple: the budget you used to burn on initial setup can finally go where projects are actually won, iterating with your business users.
ARender, our document viewing technology, is moving in the same direction, and this summer the pace has been high. Two of the four items below are about giving applications and AI a way into the viewing layer; the other two are about the viewer itself getting younger.
New APIs let applications, and AI agents, interrogate a document's content and condition: is this PDF signed? Is it encrypted? Is the file corrupted? It is the beginning of a whole new set of document introspection capabilities, precious for bulk migrations and for sharper user feedback inside the viewer itself.
The same idea that shipped for FlowerDocs, applied to viewing: an experimental MCP server that lets an AI assistant work with the documents open in ARender. Early days, deliberately. More when it matures.
Our React-based viewer keeps its rolling-release rhythm, with ARender Classic as the reference bar. Landing right now: multi-document view and the full range of ARender download options, redacted versions, annotated copies, exports to standard formats. Modernizing the viewer without giving up what made Classic trusted.
Native spreadsheet viewing joins the family: workbooks render directly in the browser, no desktop round-trip, no lossy conversion. One of our most heavily requested improvements, straight from the audit and finance departments that live in spreadsheets.
The spreadsheet viewer in motion. Early build: this is not the final rendering.
Every FlowerDocs or Uxopian AI project accumulates customization: handlers, prompts, document classes, GUI configurations, scripts. We already had a CLI for that work, but it was limited to FlowerDocs. uxc is the next step: one unique builder companion that works across our products and treats your customization as code, with an AI-based development flow and packaging built in. Pull from a server, edit locally in your own tools, push back validated and in dependency order, test the result, check for drift, package what you built.
uxc already has the primitives an ecosystem of builders needs: dependency management, multi-version handling, testing, packaging. That is the point. We want it to become the foundation people build on top of, and we are deliberately taking a multi-quarter path from experimental to mainstream, using it on our own projects first before enlarging the circle.
Experimental, but open, and harmless to try. uxc is open source on GitHub and runs entirely on your side: there is nothing to install on your servers. You do not have to wait for "mainstream" to clone it, use it, and tell us what you build.
The best proof of a platform is what gets built on it. This year, together with enterprise customers, we assembled two advanced solutions mixing FlowerDocs and Uxopian AI: one for purchase order management, one for contract management. And building them, something clicked. With the new development flow and AI coupled into the platform, the Uxopian portfolio revealed its full flavor: complete user experience overhauls, dashboards, wizards and worklists came together in days, not months, exercising the extensibility, the pluggability and the component APIs FlowerDocs has always had.
The interesting part is not that AI is involved. It is how. In these solutions, AI sits at the core and behind the scenes: transparent, but everywhere. A contract is not a PDF dropped into a chat window; it is a structured object whose parties, clauses, obligations, amounts, dates and versions are fields the platform governs and the AI works against. Prompts are versioned assets. Extractions land in metadata that workflows react to. Every AI decision leaves a trace a human can audit. Exactly what the complex, high-stakes files of a purchasing or legal department demand.
Legal AI redlining clauses, with comments, right inside the Word document.
A clause deviation queued for human decision: accept it, or send it back to negotiation.
Contract intelligence on the case file: clause-by-clause conformity, deviations and risk.
One blob of text goes in, one answer comes out. No knowledge of your data model, no memory, no workflow, no trace. Fine for a quick summary. Not for an operation you are accountable for.
Clauses, parties and obligations live as governed objects. The AI extracts, compares and drafts against them, workflows pick up the results, and every step stays traceable end to end.
One more thing, and it ships with the July release of Uxopian AI, which will get its own blog post: agentic plans are entering beta, and will mature over the coming quarter. The problem they address is one every serious AI project hits sooner or later: fixed orchestration is predictable but rigid, while fully agentic behaviour is flexible but hard to trust. Agentic plans let you mix both in a single process, with every step traced.
And agentic does not mean humans out of the loop. The first use cases we are building are around capture, and agents can hand work back to people when it matters: a human reviews what the agent extracted, accepts, rejects or corrects it directly on the document, leveraging the power of ARender and its annotations.
Autonomy where it helps, control where it counts.Agents and Plans, the new Agentic section shipping with the July release of Uxopian AI.
Human in the loop, on the document: the agent's extractions land as annotations the reviewer accepts, rejects or corrects in place.
We will keep the deep dive for its own post, once the beta has met its first real projects.
Five stops, and most of them deserve a full article: the data-model-versus-copilot argument, a uxc hands-on, the July release of Uxopian AI, agentic plans as they mature. Expect follow-up posts on many of these topics over the coming couple of months. In the meantime, if any stop on this tour resonates with your roadmap, we would love to hear about it.
Curious about the purchase order and contract solutions, or ready to put the FlowerDocs MCP Server to work on your own backlog? Let's talk.