Plugins

Extend the canvas, on your terms.

Null Space ships as a focused core, then gets out of your way. Plugins are small, self-contained bundles you install in a click to add new powers — isolating AI agents, switching Claude logins, driving your planner from an agent, or typing with your voice. Every plugin below is live today, and all of them are free for any account.

How plugins work

Power you opt into, not bloat you carry.

The app is a clean host. Each plugin is a sandboxed bundle that registers new capabilities and surfaces only when you enable it — nothing runs that you didn't ask for.

Catalog

Browse & install in-app

Open Settings → Plugins inside Null Space. Pick a plugin, hit install, and it downloads, verifies its checksum and registers itself — no terminal, no setup.

Sandboxed

Self-contained & local

A plugin is a manifest plus its own scripts. It runs on your machine through a strict JSON contract, keeps its files in its own folder, and touches nothing until enabled.

Free

Free for every account

All current plugins are free to install and update. Each declares the minimum app version it needs, so you always know it fits your build before you add it.

The catalog

Four plugins, available now.

Each one solves a real friction in agent-heavy, multi-repo development. Here's exactly what they do.

Give every AI agent its own copy of the working tree, so parallel agents never collide.

Run two agents on the same project and they fight over the same files — one's half-finished edit becomes another's broken build. This plugin removes the collision entirely by running each agent inside its own isolated git worktree, on its own branch (ns/<agent>), with its own checkout of your code.

It's built for the messy real case, not just a single clean repo. Point Null Space at a plain folder that holds several separate repositories at once and the plugin mirrors the whole launch folder: every child repo becomes a per-agent worktree, and everything that isn't a repo is symlinked through. Agents work as if they each had the full project — because they do.

  • One toggle. Turn on “Run new agents in an isolated worktree” and every agent you open after that gets its own tree automatically.
  • A diff that's actually scoped. The agent's Diff window shows only what that agent changed — repo-qualified per file, no noise from your other work.
  • Merge or discard in one action. Happy with the result? Merge it back. Not? Discard the whole worktree — your real branches were never touched.
  • Auto-cleanup. Close an agent whose worktree is clean and it's quietly removed, so isolation never leaves clutter behind.
Best paired with multiple agents on one space — let them all run at once, fearlessly.

Keep several Claude Code logins side by side and switch the active one in a single click.

If you hold more than one Claude account, hitting a rate limit usually means logging out, logging back in, and losing your place. This plugin keeps all your Claude Code logins saved together and lets you flip the live account instantly — no sign-out dance.

It also makes your limits legible. The toolbar usage gauge splits into a row per account, each with its own limit bars and reset times, so you can see at a glance which login still has head-room and switch to it before you're blocked.

  • One-tap switch. Pick an account from the gauge and the live login changes for the agents you launch next.
  • Per-account limits. Every saved login tracks its own usage windows and reset clocks, separately and in real time.
  • Phone alert on reset. Get a push the moment a paused account's limit frees up again — so you know exactly when to hop back.
  • Save the current login. Capture whatever account you're signed into right now with one button, then add as many more as you like.
The usage gauge behaves exactly as before until you install it — zero change for single-account users.

Let Claude Code read and drive the planner of the space it's working in — without ever closing a task itself.

Installing this plugin teaches Claude Code the /nullspace-tasks skill, wiring the agent straight into your space's Planner. The agent can see the tasks, see the focus files you pinned to each one, and keep its progress honest by ticking a live to-do list as it works — all reflected back in the cockpit window in real time.

Crucially, it keeps you in control. The agent can move a task to Review with a hand-off note when it thinks the work is done, but never straight to Done — that final call is always yours, via Approve or Reopen on the review card.

  • Mentioned automatically. Pin a task to an agent and the skill is invoked for you in its chat — the agent starts already knowing the task.
  • Live to-do list. The agent can build, reorder and tick off the task's checklist as it goes, and you watch it fill in on the canvas.
  • Focus files travel with the task. Whatever you attach to a task becomes the agent's focus, even when it's running inside an isolated worktree.
  • Review, never Done. A hard rule baked into the skill: the agent reports for review and you approve — it can't tick its own homework.
Pairs naturally with the planner's assign-by-thread cable and with worktree isolation.

Talk instead of type — your words land at the cursor, transcribed entirely on your Mac.

Hold or toggle a hotkey, speak, and the text appears wherever you're focused: an AI agent prompt, a task, a Markdown file or a document. A glowing waveform pill shows it's listening, and the transcript is inserted smartly into whatever card has focus — no copy-paste, no switching apps.

It's private by design. Speech-to-text runs 100% on-device with a local model you choose; no audio ever leaves the machine. On first use the plugin quietly bootstraps its own private environment and downloads the model — with a real progress bar showing bytes on disk, not a fake spinner.

  • Your choice of model. Parakeet v3 (fastest on Apple Silicon, the default) or Whisper for 99-language coverage — pick and download from the plugin's settings.
  • Toggle or push-to-talk. Two configurable chords — including modifier-only combos like ⌥⌘ — for hands-on or hands-off dictation.
  • Smart insert. The words go to the right place automatically — a terminal paste into an agent, or text inserted into the editor you're in.
  • Nothing leaves your Mac. Recording and transcription happen locally; the microphone permission lives in the app and audio is never uploaded.
Apple Silicon only — the local models run on Apple's MLX framework.
Getting started

Installed in three steps.

01

Open Settings → Plugins

Inside Null Space, head to the Plugins screen to browse the live catalog.

02

Install & enable

One click downloads the bundle, verifies it, and flips it on. Updates are one click too.

03

Use it on the canvas

The new capability appears right where it belongs — isolated agents, the usage gauge, the planner, or any text field.

Null Space

Start with the core.
Add the rest when you need it.