Security and privacy

How Windlass isolates your workspace, handles your data, and what you can do to keep your work safe.

Windlass runs code on your behalf and handles files you care about. This page explains how your workspace is protected, what happens to your data and how to use the product safely.

Isolation

Each workspace runs in its own isolated container. The agent's shell commands, the code it runs, your project files and any connectors you configure all live inside that container, separate from every other account. Nothing another customer does can reach your environment, and nothing your agent does can reach theirs.

Encryption in transit

All traffic between your browser and Windlass, and between Windlass and the services it uses, is protected with TLS. This includes the Workspace, the Files page, uploads and downloads, and the requests made to the model.

How your data is processed

When you send a message, the relevant parts of the conversation, any attached files and the file contents the agent reads are sent to the model to generate a response. Windlass uses DeepSeek's models, so those prompts and attachments are processed by DeepSeek's API.

Windlass does not train models on your content. Your prompts, files, code and session history are used only to run your sessions.

Your data stays yours

Private by default. Sessions and files are private to your account. Nobody else can see them unless you export and share them yourself.
Export any time. Download files from the Files page, download a session with /export, or export your files from Settings. You are never locked in.
Delete your account. From Settings you can delete your account. Export what you want to keep first.

Account security

You sign up with an email address and a password, and the address is verified before the account is active.
If you forget your password, use the password reset option on the sign-in page and follow the email you receive.
Use a strong, unique password. Because the agent can act inside your workspace and through your connectors, your Windlass password protects more than a chat history.

Recommendations

The agent is capable and, depending on your permission preset, may act with some autonomy. A few habits keep that power safe:

Do not paste secrets into the chat. API keys, passwords and private keys do not belong in a conversation or an attachment. If a task needs a credential, give it to a connector instead.
Use connector-scoped tokens. Create tokens with the minimum access a connector needs, such as read-only access to a single repository. Revoke them when the work is done.
Review agent actions. Read what the agent proposes before approving risky actions, especially commands that delete data, change external systems or send anything outside the workspace.
Match the permission preset to the task. Use a stricter preset when working with production systems or sensitive data.
Treat web content as untrusted. The agent can browse the web. Pages it reads are information, not instructions, and the agent is expected to treat them that way; if a result looks odd, verify before acting on it.
Clean up. Delete files you no longer need, and remove connectors and their credentials when a project ends.

Reporting a concern

If you believe you have found a security issue, or you have questions about how your data is handled, email [support email]. Include enough detail to reproduce the issue, and leave secrets and other people's personal data out of the report.