Privacy Notice
Privacy Notice
This notice describes the information Sulcus currently processes to authenticate accounts, operate projects and runs, and show run history. It reflects the current early-access product and will change as the product does.
Last updated: October 19, 20261. Information we process
When you use Sulcus, the product processes:
- Account information: a Firebase user identifier and, when provided by the sign-in flow, an email address, its verification status and a display name. Sulcus also keeps the user, workspace, membership, role and timestamp records needed to authorize access.
- Project configuration: project name and description and, for hosted projects, the public repository URL and Git reference, the Python file to run, the framework, dependency mode and network setting.
- Run information: workspace and project associations, name, configuration snapshot, status, timestamps, token-usage totals and limits when available, and sanitized error information.
- Runtime activity: ordered runtime events from the supported integrations, such as lifecycle, agent, model-call, tool-call and token-usage activity.
- Hosted application output: lines a hosted run writes to stdout and stderr, up to a limit, and diagnostics needed to show failures. This can include any content your code prints.
- Paired computers: a name and platform for each computer you pair, a credential for it, and when it was last seen.
- Local agent activity from Claude Code and Codex: identifiers, agent and tool names, status, timing, counts and token usage, together with the project folder’s name, a hash of its path and its Git remote address without credentials.
- Approvals: the tool or action an agent asked permission for, your decision, and when it was made. For Claude Code and Codex on a computer set to share decision context, an approval also carries the command, working directory, affected file paths and, for a network request, the hostname.
- Telemetry from connected services, such as a Google ADK service or a Microsoft Copilot Studio environment: agent, tool and model names, identifiers, timing, error status and, where reported, token counts.
- Connection credentials and settings: for a service you connect, the identifiers that locate it and the token or credential the connection needs. A Google ADK ingestion token is stored as a hash. A Microsoft Copilot Studio connection stores your tenant, environment, Application Insights and client IDs, and the client secret in encrypted form.
If you contact support, we receive the information you choose to send in that message.
2. How we use information
Sulcus uses this information to authenticate users, authorize workspace access, create and execute projects and runs, display run status and history, apply token limits and approval decisions, validate inputs, troubleshoot failures, protect the service, and respond to support requests.
3. Authentication and service providers
Sulcus uses Firebase Authentication, a Google service, for sign-in with Google, GitHub or email, and for server-side identity verification. Application data such as projects, runs and events is stored in a database operated by Sulcus. A complete list of service providers has not been published yet; contact us if you need it.
4. Repository and Run data
For hosted runs, Sulcus accepts public HTTPS Git repositories without credentials in the URL. Private repository authentication is not supported. Sulcus retrieves the repository content for the selected reference, installs dependencies, runs the configured Python file, and collects the runtime events, application output and diagnostics described above. A run keeps a snapshot of its configuration so later project edits do not rewrite its history.
The framework integrations are built to record which calls ran, not their content: they do not collect prompt text, model responses or tool inputs and outputs. Application output is different. Whatever a hosted run prints is stored with the run, so do not print information you do not want kept.
For Claude Code and Codex on your computer, prompts, model responses, command output, file contents and file diffs are not sent to Sulcus. Each paired computer has a sharing level that you set on that computer with sulcus device privacy; Sulcus cannot change it. A computer that is new to Sulcus starts at decision context: when an agent asks for approval, the request also carries the command (up to 2,000 characters), the working directory, up to 10 affected file paths and, for a network request, the hostname. This is kept on the approval record, shown to the workspace members who can decide it, and not added to the run’s events. Common credential shapes are removed on a best-effort basis, so a command or path can still contain a secret or other sensitive text. At the private level, used by computers set up before sharing levels existed and by sulcus login --private, an approval shows only which kind of action is waiting, and command text and file paths are not sent. If you use the optional feature that starts an agent on a paired computer from Sulcus, the task you type is sent through Sulcus and kept until the agent starts or the request ends.
For a connected Google ADK service, prompts, responses, tool arguments and results, and user identifiers present in the telemetry are discarded on arrival.
For a connected Microsoft Copilot Studio environment, Sulcus reads telemetry from the Azure Application Insights resource you designate. It requests identifiers, timing, outcome, and agent, tool and model names. It does not request message content, tool inputs or results, or user identity.
5. Secrets and sensitive values
A credential saved for a connected service, such as a Copilot Studio client secret, is stored encrypted until you disconnect that service. API keys and other secret values for a hosted run are different: they are supplied for that run only. They are not saved with the project, the run or its events, and Sulcus does not offer a secret vault for runs. Runtime events, application output and diagnostics pass through best-effort redaction for supplied values and common credential shapes. Redaction is not guaranteed, does not remove other sensitive content, and cannot prevent code you run from sending a value elsewhere.
6. Data retention
Sulcus stores account, workspace, project and run information, together with a bounded event history, as needed to operate the service. Event history is limited to the 5,000 most recent events per run by default. The product has no published time-based retention period or automatic deletion schedule; retention and deletion controls are still being built. Projects with historical runs cannot currently be deleted in the app. To ask about deleting your data, contact us.
7. Security
Sulcus uses authenticated API access, workspace-membership authorization, repository validation, best-effort redaction, and container and resource controls for hosted runs. Hosted runs are intended for repositories you trust; the isolation is not presented as a guarantee. See Security and data handling for details.
8. Your choices and contact
For privacy questions or requests about your data, contact support@sulcus.dev.