Nimbalyst Beta

Nimbalyst shows the agent's proposed changes next to your original text, so you see exactly what will change before accepting it. You can approve, reject, or redirect the agent mid-task. Why does that matter? Because context stays intact when the agent reads your project structure and related files instead of just the open document.

You edit markdown visually instead of raw syntax. Formatting, tables, and links render as you type, while the underlying files stay plain markdown sitting on your disk exactly where you expect them, so your docs keep working in git and any other editor you already use without any lock-in to a proprietary format.

The session manager runs agents in parallel and you organize them on a kanban board by phase: backlog, planning, implementing, complete. A sidebar lists every file a session reads or changes, and you can open them or manage git state from there. AI commits are built in.

The workspace includes purpose-built editors for mockups, Excalidraw diagrams, mind maps, data models, slides, spreadsheets, and CSV files. You work across all of them without switching apps or reconfiguring your agent.

Tasks live inside the workspace, not in a separate tracker. Agents read your backlog, pull context from related items, and update the board as they finish work, which means the ticket you see always reflects what the agent actually did on disk. Dragging an item updates what agents see, so there's no syncing and no stale tickets.

On the Teams plan, teammates and their local agents edit the same shared docs, diagrams, and trackers with live cursors. You promote any local file to a shared one in a click and review its full change history. Team chat sits in the same workspace.

The desktop and iOS apps are MIT licensed, which means you can read the full source code, file issues, build your own extensions, or fork the whole thing and keep it running the way your team wants. An extension system lets you build custom visual editors, and those extensions inherit multiplayer support automatically.

Nimbalyst connects to the Claude Code or Codex subscription you already have, an Anthropic or OpenAI API key, or local models through LM Studio. It supports an MCP server for additional tooling, so you shape the environment around the models you trust.