Claude Code,
in your notch.

Q: What does it change in my config, and how do I remove it?

It adds its hooks to ~/.claude/settings.json, and with plan limits on, its status line, after a backup in ~/Library/Application Support/Gnista/Backups/. Yours stay. Settings ▸ Agents ▸ Claude Code ▸ Disconnect removes exactly Gnista’s hooks and puts your status line back as it was.

Claude Code notifications, permission approvals and usage limits on your Mac

Gnista shows every Claude Code session in your MacBook’s notch: working, done, or waiting for you. Allow its tool calls, answer its plans and questions, and reply when a turn is done, without going back to the terminal, the editor or the Claude app.

Try free for 7 days No account, no card · then €19 once

01At a glance

Gnista with Claude Code, at a glance.

Where it runs: Any terminal, any editor’s terminal, the Claude Code extension in VS Code, Cursor and Devin, Code sessions in the Claude app, Claude in Xcode (how), and Claude Code on a Linux server, over SSH.
Setup: Hooks in ~/.claude/settings.json, added after a backup. Nothing to trust in Claude.
Needs you: Permission requests, plans and questions, the moment Claude asks.
Answer from the notch: Allow, Always (Claude’s own suggested rule, spelled out), or Deny with an optional reason Claude reads; plans and questions too.
Reply from the notch: Yes, wherever Claude runs: a hook wakes the session with your words as they are.
In the row: Claude’s title for the session, your prompt, the latest message, the tool in use, task progress, the permission mode, context and cost.
Subagents: How many are done, the one in front and its step; background subagents stay on the row until they finish.
Cost and limits: Cost at API list prices, and your 5-hour and weekly limits once you switch on Gnista’s status line.

Approve without the alt-tab.

The notch opens with the whole request: the command, or the diff with line numbers, folded to the hunks that changed. Allow waits half a second, so a stray click can’t say yes. Always applies the rule Claude suggests, shown in plain words first. ⌘Y allows, ⌘N denies. Claude’s own prompt shows too: answer either one.

Plans and questions, answered in place.

When Claude leaves plan mode, read the plan in the notch and choose Approve, Accept edits, or Keep planning with feedback. Questions come one at a time, with Other… for an answer in your own words.

One more thing? Reply.

When a turn is done, type a reply in the notch. In Terminal, iTerm, Ghostty, a tmux pane, or an editor’s terminal with Gnista’s extension, Gnista types it in. Anywhere else, in the VS Code extension or the Claude app, a hook Gnista adds wakes the session with your words as they are. claude -p runs aren’t held: they finish as usual. Quit Claude Code in its terminal already? A reply resumes the session in the background, and anything that needs your yes waits for Allow or Deny in the notch, with Answer prompts in Gnista on.

Subagents, on one line.

While Claude hands work to subagents, the row reads like “1 of 2 done · Review the API · Reading api.ts”. Subagents running in the background stay on the row after the reply, until they finish or you stop them.

Know where your limits are.

Switch on Claude Code plan limits and your 5-hour and weekly windows sit under your sessions. They come from Claude Code’s own status line: Gnista wraps yours if you have one, and puts it back exactly when you switch off. No Keychain, no sign-in, no network. Cost is at API list prices; for a model Gnista doesn’t know, describe it in a model file.

What it can’t do with Claude Code, and why.

Cowork sessions in the Claude app run in a virtual machine, so no hooks reach your Mac: their rows come from the app’s own session files, and you answer their requests in the app.
In Terminal, iTerm, Ghostty, a tmux pane, or an editor’s terminal with Gnista’s extension, Gnista types your reply in. Everywhere else it goes through a hook: Claude’s window shows it as “Stop hook feedback”, not as your message, and it works for an hour after the turn ends.
Plan limits come only through Claude Code’s status line, so they need Gnista’s status line switched on.
A session that was already running when you connected Gnista keeps its old hooks. Start a new one.

02Setup

Connected in a minute.

Open Gnista and tick Claude Code. Gnista backs up ~/.claude/settings.json to ~/Library/Application Support/Gnista/Backups/, then adds its hooks next to yours, keeping your formatting.
To answer requests from the notch, turn on Settings ▸ General ▸ Answer prompts in Gnista.
For cost and limits, pick Cost, Plan limits or Both under Settings ▸ General ▸ Usage. Plan limits also need the Claude Code plan limits switch, which joins Claude Code’s status line.
Start a new Claude Code session, in a terminal, your editor or the Claude app.

03FAQ

Claude Code, asked and answered.

Can I approve Claude Code’s tool calls from the notch?

Yes: Allow, Always or Deny, with an optional reason for Claude. Turn on Settings ▸ General ▸ Answer prompts in Gnista. Claude’s own prompt shows at the same time; answer either one, and the other goes away.

Does it work with the VS Code extension and the Claude app?

Yes. Both run Claude Code with your hooks, so you get the row, approvals, plans, questions and replies, and a click opens the session’s tab or the session in the Claude app. VS Code asks once before it lets Gnista open the link. Cowork sessions show too, read from the app’s own session files: a click opens one in the app, where you answer its requests.

Why doesn’t my session show up?

Claude Code reads its hooks when a session starts, so a session that was running before you connected Gnista doesn’t have them. Start a new one. Settings ▸ Agents ▸ Claude Code shows what Gnista added.

Does Gnista read my Anthropic credentials?

No. Plan limits come from the status line Claude Code itself runs. Gnista never reads your Keychain or calls Anthropic.

What does it change in my config, and how do I remove it?

It adds its hooks to ~/.claude/settings.json, and with plan limits on, its status line, after a backup in ~/Library/Application Support/Gnista/Backups/. Yours stay. Settings ▸ Agents ▸ Claude Code ▸ Disconnect removes exactly Gnista’s hooks and puts your status line back as it was.

Does it work with Claude Code on a server, over SSH?

Yes, on a Linux server. Connect the server in Settings ▸ Remote, with a host from your ssh config: Gnista puts its helper there and adds its hooks to Claude Code’s settings on the server. Its sessions show with the machine’s name, its title, latest message and cost; answers from the notch reach Claude there, and a click brings up the terminal tab you ssh’d in from, tmux pane included, or your editor’s Remote-SSH window. Step by step: agents over SSH.

Gnista also keeps an eye on

Wrote your own? It can report to Gnista with one line of JSON: custom agents.