Sightdeck AIAgentic Studio

User Guide

Permissions & privacy

macOS asks before an app touches anything sensitive. Here is every prompt Sightdeck AI can trigger, why it appears, what the app talks to on the network, and what never leaves your Mac.

The short version

Sightdeck AI only requests what it needs to launch and watch agent sessions for you. Every prompt below is tied to a feature you actually use. The app does not read your music, photos, camera, microphone or location, and it has no analytics or tracking. Your tasks live in plain files — in the project or on your Mac, as you chose when you added the project.

Two different kinds of permission

This page covers what the app asks macOS for. There is a second, separate question: what the agent is allowed to do inside your project once a task starts.

That one is a switch on each provider — Full Access Without Confirmations for Claude, Danger Full Access for Codex — and it is off by default. It is worth a conscious decision before your first real task: see Full access for what it covers and where to find it.

Prompts you may see — and why

macOS shows these the first time the matching feature runs, not all at once on launch. Grant the ones for features you use; you can change any of them later in System Settings.

  • Control “Terminal” (Automation / Apple Events). Needed only when sessions run in Terminal windows. By default a project runs its agents embedded in the Sightdeck window, and then no Terminal is involved. If you switch Run agents embedded in Sightdeck off in a project's General settings, Sightdeck opens each session in a Terminal window and needs to find, focus and close those windows again. The prompt reads roughly “Sightdeck AI wants access to control Terminal”.
  • Keychain access for “Claude Code-credentials”. To show your Claude usage limits, Sightdeck reads the sign-in Claude Code already stored in your keychain and asks Anthropic for your 5-hour and weekly usage. macOS may ask whether to allow this. Decline it and everything else still works — only the usage meters stay empty.
  • Notifications. When a task finishes, Sightdeck AI can post a banner so you do not have to keep the window in focus. Appears the first time a task completes. Decline it and everything still works — you just lose the done banner.
  • Files & folders. If your project or task files live under ~/Documents, ~/Downloads or ~/Desktop, macOS may ask before Sightdeck AI watches them for changes. Needed so the board updates live when a task changes.

What Sightdeck AI does not access

Some Mac apps ask for things they never actually use. Sightdeck AI shuts those off at launch, so you should not see prompts for:

  • Apple Music and your media library. Sightdeck AI has no audio or media feature. An older build could show this prompt by mistake; the latest build no longer does. If you granted it on an old version, it was harmless — the app never touched your library, and you can revoke it under System Settings → Privacy & Security → Media & Apple Music.
  • Local network discovery. Sightdeck AI never scans your local network or discovers devices on it. The only local-network feature is Sightdeck Remote, which serves a page to phones you pair yourself.
  • Camera, microphone, photos, location, contacts, calendar. No feature touches any of these.

What stays on your Mac

  • Your tasks, concepts, attachments and settings. Settings marked as staying on this computer — such as the provider a project runs on — are kept in the project's local configuration and are not shared with your team.
  • Your provider profiles. Sightdeck passes no credentials to your agents; Claude Code and Codex each use their own sign-in.
  • Token costs and Insights are calculated from the session records on your Mac.
  • The Profiler recording of each run. You can switch recording off under Settings → Insights.
  • Your license. The key is checked on your Mac; activation needs no connection.

Your code does go where your agent sends it: Claude Code and Codex talk to their own services, under the terms you agreed to with them.

Network connections

Sightdeck AI sends no telemetry. The app itself talks to the network only for these:

  • Update check. A periodic check against the Sightdeck website to tell you when a new version is out.
  • Model list. A list of current models and their prices, fetched from the Sightdeck website so model pickers and costs stay current.
  • Claude usage limits. Your own 5-hour and weekly usage, fetched from Anthropic with Claude Code's sign-in — no task content is sent.
  • Store. Browsing, installing, rating and publishing templates in the Store. Ratings and comments are tied to an identifier derived from your license, never your license key or email address.
  • Help chat. Questions you ask the Help chat go to the Sightdeck website to be answered from the documentation. The conversation is kept on your computer until you delete it.
  • Your integrations. Jira, GitHub, webhooks and Sightdeck Anywhere connect only where and when you set them up.

Manage permissions in System Settings

  1. Open System Settings → Privacy & Security.
  2. Pick the category — e.g. Automation for the Terminal control grant, Notifications, or Files & Folders.
  3. Toggle Sightdeck AI on or off. Turning off Automation stops sessions that run in Terminal windows until you re-enable it; embedded sessions are not affected.
Sightdeck AI setup steps with provider, agent and skill
The setup steps — provider, agent and skill in one place.