Give Your AI Coding Agent a Visual Workspace

Coding agents are good at files and bad at context: they can’t see the whiteboard photo of your architecture, the product decision from last week, or which of five approaches you actually picked. The fix isn’t a bigger prompt — it’s a shared workspace where planning happens in files the agent already reads.

The workspace is the interface

Point the agent at a Toxvas folder and it sees what you see: cards as Markdown files, boards as JSON layout, attachments beside them, and a spec file documenting the format. No API key, no MCP server to configure, no account — any agent that can run a shell reads with show and search, writes with new, append, and link. JSON out by default, exit codes it can branch on, --dry-run on every write so it can rehearse safely.

Plan the feature on the board first

Sketch the approaches as cards — “migrate auth”, “cache layer”, “do nothing” — link them with labeled connectors (“blocked by”, “cheaper than”), and frame the decision. Then tell the agent to read the board: it picks up the decision and its reasoning, because both are files. When it writes the implementation notes back, they appear on your open board within about two seconds. The plan and the record stop being two different artifacts.

Specs that survive the session

Agent sessions evaporate; folders don’t. Decisions land as Markdown with frontmatter (id, dates, tags), tasks as - [ ] do thing @due(2026-11-04), references as [[other card]]. Next week — or the next agent, or a different model entirely — opens the same folder and continues from files, not from a chat log you have to paste. doctor verifies the workspace is readable and writable before the agent touches anything.

The honest limits

This is planning and memory, not execution: the agent can’t run your code from here or see your repo unless you attach the relevant files. And it’s one Mac, one folder — no shared team workspace, no real-time collaboration. For solo developers who plan before prompting, though, it closes the loop other tools leave open: the board you thought on is the folder your agent works in. Try it free for 7 days, point your agent at the folder, and watch the canvas follow.