Privacy
Understand the difference between usage analytics, stored run history and data on your computer.
Sculptured stores team designs and run history so you can run agents and review their work. In the hosted beta, that operational data is stored in your account's Sculptured instance. In local mode, it is stored on your computer.
The hosted beta also collects usage analytics: feature use, team structure, run statistics, token totals and errors. Analytics exclude prompts, messages, jobs, agent output, file names and file contents. Self-hosted Sculptured does not collect usage analytics.
Excluded from analytics
- Prompts, messages and jobs
- Agent output and transcripts (for analytics)
- Team names and agent names
- Folder names and paths
- File contents
- Check commands and their output
Analytics events
| Event | From | What it says |
|---|---|---|
auth.signin | Sculptured site | You signed in, and how (password, Google or GitHub) |
invite.used | Sculptured site | An invite was used to join |
device.paired | Sculptured site | A computer was connected, and its operating system |
device.revoked | Sculptured site | A computer was disconnected |
api | Sculptured site | A request: its kind, the page or action with names removed (for example /api/teams/:id/git), its status and how long it took |
session.start | Your browser | You opened Sculptured |
session.end | Your browser | You left |
page.active | Your browser | Every 15 seconds while the tab is in front: which kind of page (names removed) |
ui.action | Your browser | You used a feature such as Describe it, Token use or Download run (its name only) |
team.created | Your Sculptured | A team was made, and from what (a template, a file, a description) |
team.saved | Your Sculptured | A team's shape (counts only: agents by kind, groups, arrows; whether checks, limits, copies or a workflow are used), at most every 10 minutes |
team.started | Your Sculptured | A team started: how many agents, which programs, how many models, its shape |
team.stopped | Your Sculptured | A team stopped |
run.summary | Your Sculptured | A finished run: duration, tokens in and out, agents, spawned agents, failed agents |
check.ran | Your Sculptured | A check ran, and whether it passed (never the command or its output) |
copies.enabled | Your Sculptured | Agents' own copies were turned on |
plan.held | Your Sculptured | A start waited because a plan was near its limit |
Analytics events are kept for 180 days. After that, only daily totals remain.
Data stored to run your teams
Analytics are separate from the data Sculptured needs to run your teams. With an account:
flowchart LR
subgraph PC["Your computer"]
F[("Your files and repositories")]
C["Claude Code / Codex sign-in"]
A["Agents"]
end
subgraph SV["Your Sculptured on the server"]
T[("Teams, runs, messages, transcripts")]
end
subgraph AN["Analytics"]
E[("Events in the table above")]
end
A -- "events: replies, tool calls, results" --> T
T -- "counts and shapes only" --> E
F -. "only what you ask to see: folder lists, Git status, diffs" .-> T
| Data | Where | Why |
|---|---|---|
| Your teams' designs, including jobs | Your own Sculptured on the server | To show and run your teams. |
| Messages, agent replies, tool calls and results, questions and answers | Your own Sculptured on the server | So you can watch a run, inspect its evidence and come back to its history. |
| Your files and repositories | On your computer | Agents work there. Requested folder listings, Git status and diffs can reach the server. File content may also appear in captured tool results and agent replies. |
| Claude Code and Codex sign-in | Only on your computer | Sculptured never reads or stores them. |
| Your name, email, sign-in methods, and a password hash if you use a password | The account database | To sign you in. |
| Your computers: name, operating system, last seen | The account database | For the Computers page. |
The server's data is backed up every night. Admins can see account details, computers and analytics. They cannot open your teams, runs or files.
What your agents send
Claude Code and Codex send your agents' prompts and the files they read to Anthropic or OpenAI, as they do when you use them directly. That traffic goes from your computer to the provider, under your own Claude or ChatGPT plan or API key. Sculptured is not a party to it.
Traces you choose to send
If you turn on Send traces for a team, its runs go to the address you chose. By default that is timings, token counts and costs, with names: the team's, its agents', the computer's id, the models and the tools used, and the reasons for budget changes and suspected stalls. Only if you choose Also prompts and outputs do texts go too (each cut to 4 KB): prompts, agents' replies, and tool inputs and outputs, which can include file contents and command output. The API key you give is kept on the server and sent only to that address.
Community
Community uses your Sculptured account. Publishing a block makes its definition public: its agents' roles and jobs, its connections, and the title and license you choose. Machine-specific folders, model overrides and workflow input are removed first, but review the jobs for private details before you publish.
Your profile is public too: your name, handle, bio, links, picture, when you joined, your published blocks and their counts. Empty fields are not shown. Reports can be about a block or a profile. See Community.
Your choices
- Run it yourself to keep everything on one computer with no analytics. See Run it yourself.
- Export a team as a file at any time. See Team files.
- Disconnect a computer from the Computers page. Its key stops working at once.