Lemni

PDF drawing sets and Word documents become compact cards you page through in place; spreadsheets and CSV points lists become tables. Files are kept inside the board, so the job's paperwork travels with it.

Pen, sticky notes, arrows and shapes work anywhere, including right on top of a drawing, so markup and questions stay with the sheet they are about.

Pages sit in categories in the top bar, and a category's kind decides which tools apply: documents, UI pages for graphics, or the generated wire sheet. You can add your own, like Commissioning, that work like either.

Past the core canvas, every feature is a plugin you can switch off: documents, pen and shapes, the libraries, charts and tables, behaviors, links, web views, the station connection and the AI connection. Niagara through BASidekick Stream is the first data connection; others are intended, and being able to write your own plugins is the plan.

Libraries come from plugins too. The symbol library is HVAC and BAS equipment with ports, flips and live states. The data library holds trend charts, stat tiles, gauges and tables, and dragging a point in from the station offers the right one.

Drop a station point on any shape to make it live: symbol state, color, level, labels, rotation, motion or flow. Present mode hides the editor and leaves a working graphic.

You can link pretty much anything to anything: a device or point in the station, one page of a drawing set, a document, a frame on a graphic, a whole page. Each link carries whatever context you want to give it, in your own words, with no structure required. The wire sheet is generated from those links and shows how everything connects, with each comment on its wire. Tags give an AI a little context about a point; this gives it, and the next person on the job, the rest of the story: which drawing a device comes from, which graphic shows it, and why. None of it needs a visual: a link can be nothing more than context for an AI. But I think a hybrid interface like this, where the same information is equally there for the person, will be the norm for a while.

I'm aiming for small things that add up and an interface that's intuitive throughout. Drag a point out of the station onto the board and a menu asks what it should become: a chart, tile, gauge or table; a symbol that already animates from it (a fan for a fan status, a damper for its position); or a control. Drop a whole equipment folder and you get a panel or a table of it. Drop a point on a shape and it's ready to drive it. Each plugin's options can be hidden from that menu if you don't use them.

There is no built-in assistant and no API key held by the app. Lemni runs an MCP server, so the AI app you already pay for (Claude Code, Claude Desktop, Codex) connects to the open board. Plugins contribute its tools, so it can read documents, build and lay out graphics, bind points and link things, and potentially help with controlling data: any write to the station waits for Allow once and an are-you-sure in the board, and is audited. It uses what you already have, works with any MCP app, and does nothing until you connect it.