A local-first visual workspace: an infinite canvas of Markdown cards, in a folder you own.
Cards, connectors, mindmaps and colorable sections on an endless board, plus images and freehand ink. Cluster what belongs together, link what relates, circle what matters, and watch the shape of the problem appear.
Open the AI dock and bring your own key: Claude, GPT, Gemini, or fully local with Ollama. It reads your board, then makes real edits: new cards, connections, sections. Card rewrites arrive as word-level diffs you approve, and every run is one undo away.
Cards are plain Markdown, boards readable JSON, so Claude Code, Cursor or Codex connect in one click. With 27 MCP tools, agents search your graph, write cards, wire connections and frame whole sections. Their edits land on your open board live, merged with yours, never clobbered.
Point an agent at a wall-of-text note and ask for order. It drafts a plan: new cards pulled from the text, linked back to their source, grouped and connected, never raw coordinates. You review the draft first, down to an ASCII sketch in the terminal; nothing lands until you approve. Then a deterministic engine places all of it, or none.
Every agent edit is recorded: who changed what, and when, with a word-level diff. Come back to a "while you were away" digest, review each session, and revert any edit in one click. Agents do the work. You stay editor-in-chief.
Every checkbox in every card is already a task. The Tasks view gathers them into one kanban (Open, In Progress, Done) with your boards as the swimlanes. Type @due(date) right in the line and overdue work floats up. The same note can sit on your kanban and your research map at once.
No tiers, no lock-in. Point Acres at a folder and start thinking: your notes stay plain Markdown you own.
