Privacy & Telemetry
Blueprint AI is off-by-default for telemetry. Nothing leaves your machine until you explicitly opt in, and even after you opt in the data is anonymous error reporting only — never source code, never chat content, never asset data.
Opt-in only
Section titled “Opt-in only”There are two places to flip the consent:
- The login screen has a checkbox labelled Send anonymous error reports. It’s pre-checked as a default suggestion — but the value is only committed when you click Connect (or you toggle the checkbox manually). Closing the login screen without connecting leaves consent in the “not yet decided” state.
- Editor Preferences → Blueprint AI → Telemetry → Send anonymous error reports — flip it any time, no editor restart required.
Until you’ve made an explicit choice — either ticking the box and clicking Connect, or flipping the Preferences toggle — telemetry stays inert regardless of the checkbox state. The pre-checked default never silently counts as consent.
What gets sent
Section titled “What gets sent”When telemetry is on:
- Exception type — e.g., “authentication failed”, “CLI not found”
- Stack trace — function names from Blueprint AI’s own modules
- Plugin version — to filter for regressions
- Engine version — UE 5.7, 5.8, etc.
- OS — Windows / macOS / Linux + major version
- Opaque install ID — a random GUID generated on first opt-in. Used by Sentry to group repeated events from the same install without identifying you.
That’s the entire payload. It’s similar in scope to a typical crash reporter.
What is never sent
Section titled “What is never sent”- Source code from your project
- File paths outside the plugin directory
- Chat messages, attachments, asset names, prompt content
- Account email, API keys, OAuth tokens
- Project name,
.uprojectcontents - Hardware identifiers (CPU model, MAC address, machine name)
- Anything from your editor’s project list
The telemetry layer literally doesn’t have access to any of these — it only sees structured error records emitted by the panel’s own diagnostic code.
Revoking consent
Section titled “Revoking consent”Flip the checkbox off. Two things happen:
- The Sentry client stops sending events immediately.
- The stored install ID is cleared. If you opt back in later, a new install ID is generated, so historical events from your old ID are not re-associated.
There is no “delete my data” call to Sentry from the plugin — if you want prior reports removed, send a request via the Support form with the install ID printed in Editor Preferences → Blueprint AI → Telemetry.
Account data
Section titled “Account data”Account login uses the standard Claude OAuth flow — tokens are stored by the Claude CLI in ~/.claude.json. Blueprint AI does not duplicate them anywhere except the optional locker at ~/.claude/BlueprintAI/<email>/ (Multi-Account Login), and that copy never leaves your machine.
The plugin does not maintain a Blueprint-AI-side account or any server-side record of who’s using it.
Session data
Section titled “Session data”Conversations are stored on disk as JSONL under ~/.claude/projects/<project-key>/ (or the equivalent location for your AI service). They:
- Stay on your machine
- Are not synced to Anthropic / OpenAI / GitHub except by the underlying AI client’s own protocols (the API calls themselves)
- Are gitignored by convention — don’t commit them
- Can be deleted at any time via the SessionDrawer × button or by removing the file
The plugin never reads JSONL files except for the active project’s session list and the file you’re currently viewing.
Network traffic
Section titled “Network traffic”Outbound traffic during normal operation:
| Destination | What for | Required? |
|---|---|---|
api.anthropic.com | Claude API calls — the normal AI conversation | Yes (when chatting) |
claude.ai | OAuth login redirect | Once per sign-in |
sentry.io | Error telemetry | Only with opt-in |
No connections beyond the above. Verify with your firewall if you need to.
What about the AI itself?
Section titled “What about the AI itself?”Claude has its own privacy policy. Blueprint AI doesn’t change what Claude sees: it sends the same prompts and tool results you’d see in a normal Claude session.
Read Anthropic’s terms separately. As of writing, Anthropic / Claude does not use API conversations to train models by default. See anthropic.com/legal/privacy.
If you drive Blueprint AI from another CLI-capable AI via the bring-your-own path, read that service’s privacy policy separately.
Blueprint AI is the editor-side adapter, not the AI. It can’t make a service more or less private than that service already is.