Usage data and privacy

Studio asks before it sends anything about how you use it. There are three separate choices, and the agent.

Usage data

When you first open a build that collects usage data, it asks Share usage data with Urbium?: which tools you use, agent turns and their cost, edits applied and undone, rule checks and errors, with your firm's name. Share or Don't share.

What is sent, as events:

• the app starting and ending, a tool used, a message sent (not its text);
• each agent turn's model, tokens, cost, latency, whether it failed and how many tools it called (no input or output text);
• an edit applied or undone, a rule check (the zone and how many rules failed), an export (the format), an error (its kind).

Each carries an install ID, your firm, the app version and the release channel. Never your prompts, your models, geometry or addresses. Usage data is linked to your name and email (entered in Settings) so Urbium can follow up with you. It goes to PostHog, our analytics service. Events wait in a queue in Studio's data folder and are sent in batches every 30 seconds; a failed send is kept and tried again. After you change your name or email, the next send first links your install to them.

To see or change it:

• Settings… › Usage data: Share usage data with Urbium on or off, and your Name, Email and Firm.
• Help › View Telemetry Log… lists exactly what is waiting and what was sent recently, newest first.

A build made without usage data never asks and sends nothing; Settings then has no Usage data section.

Crash reports

A separate question at first launch: Send crash reports? Send or Don't Send. Change it with Send crash reports in Settings.

When on, Studio sends to Sentry, our error service:

• a crash, with the app version, the operating system and the graphics card;
• an error in the model engine, with its traceback;
• a trace of each agent turn: the model, token usage, and each tool call's name and status. A failed call carries only its error code: tool arguments and output are never sent.

Your home folder and project paths are removed from reports, and the computer's name is dropped. Never your projects, prompts or file paths. When the agent runs on a firm key saved in Studio, Claude Code's own timing traces go too (each model request's latency, retries and status), with prompts, responses and tool contents left out. Signed in with your own Claude account, Claude Code would add your email to them, so it sends none.

Project data

The project picker has Share project data to improve Urbium, per project and off unless you turn it on. When on, in a build set up to collect it, Studio uploads the project's change log, its model snapshots and its conversations (your messages, the agent's replies, attached images and traces) when you open another project or quit. A build not set up for it uploads nothing, even with the box ticked.

The agent

The agent in the chat is Claude Code, under your Claude account or your firm's key. Your messages and the parts of the model the agent reads go to Anthropic as with any Claude Code session, under that account's terms. Studio adds nothing to that.