Vault

Add a Vault item

1. Open Vault from the company navigation and choose Add item.
2. Choose Login, API key, or Secure note. Add a clear title and the fields that belong to that type. Login and API-key items can also carry a website and private context.
3. For a login, paste an existing password or choose Generate strong password. Copy a generated password before saving when you also need to enter it somewhere outside Genosyn.
4. Choose whether every Member in the company may view the item or only selected Members may. Save it, then open Access to add people or AI Employees.

The list and detail screens show the title, username, website, type, and private context only to Members who can access the item. The stored password, API key, or secure-note body stays masked until someone explicitly chooses Reveal or Copy. Editing an item never loads the current stored value into the form; leave the value blank to keep it, or enter a replacement to rotate it.

Member access

A Vault item has either Everyone in the company visibility or Only selected Members visibility. Restricted items are absent from other Members' lists instead of advertising that a hidden credential exists. The creator and company owners or admins can always manage sharing and deletion.

View

Open the item, read its encrypted metadata and private context, and deliberately reveal or copy the stored value.

Edit

Everything in View, plus change fields and replace the stored value. It does not confer sharing or deletion control.

Company-wide visibility affects Members only. It never gives an AI Employee access. Every AI Employee starts with no Vault access and needs a separate item-level Grant.

AI Employee Grants

Open a Vault item's Access panel and add only the AI Employees that need it. Grants are independent per item, so access to one GitHub login does not expose another login or anything else in the company Vault.

Use

Discover safe item metadata. For a Login, use server-side Browser autofill for its username or password without returning plaintext to the model. A stored password can go only into a password input; API-key values and secure-note bodies have no AI plaintext-read or Browser-fill path.

Manage

Everything in Use, plus update a login's title, username, and private context. The saved website origin cannot be rebound by an AI Employee. Manage also cannot reveal, rotate, or delete the stored password. A login an employee creates receives this level automatically.

With Manage, update_vault_login can change the title, username, or private context while preserving both the encrypted password and saved website origin. Website rebinding, password rotation, and deletion remain deliberate Member actions in the Vault.

Sign in without showing the model a password

With the built-in Browser enabled, an AI Employee can complete a login without asking a Member to paste a password into Chat:

1. list_vault_items returns only granted item ids and safe metadata. It never returns the password, API key, or secure-note body.
2. The employee opens the website saved on the login item.
3. browser_fill_vault asks the App to resolve a granted username or password and fill the selected field directly in App-owned Chromium. The top page and target frame must both match the item's exact saved origin — scheme, host, and port. A password is accepted only from a Login item and only into an input with type=password.

Browser access and the employee's host allow list remain independent gates: a Vault Grant cannot enable the Browser or widen its browsing policy. The plaintext exists only at the server-side credential-to-browser boundary. It is not serialized into the tool response, model context, Run transcript, audit detail, or log. Captchas and 2FA can still use the Browser's human take-over flow.

Browser snapshots redact password-input values, including values inside frames, before the model sees them. After the session has observed or filled a password, model-requested screenshots are refused; use the redacted structural snapshot instead.

Let an AI Employee create a login safely

AI Employees can store new credentials without first learning the password in model context. create_vault_login generates a strong password inside Genosyn, encrypts it immediately, creates a company-visible login, and gives the creating employee a Manage Grant. Members can therefore recover the credential, while other AI Employees still receive no access. The employee can then use browser_fill_vault to enter that generated value into a signup or password-change form.

When a website or browser flow already put a password into an input, browser_save_vault_login can request capture from a same-origin password input. Capture always needs approval from a company owner or admin, even when ordinary Browser form submissions do not require approval. Once approved, the App saves a restricted Vault item bound to the exact current origin and gives the employee a Manage Grant. The password is not read back or included in model output. Other Members do not see the restricted item until an owner or admin changes its visibility or Member access.

Reveal, copy, and audit

Revealing and copying are separate, explicit Member actions. Each one writes a company audit event identifying the Member, Vault item, action, and time, without recording the secret itself. Creating, updating, deleting, sharing, granting, and AI use are audited too. Review the history under Settings → Audit log when investigating access or rotating a credential.

• Do not paste a Vault value into Chat; Grant the item and use a governed action.
• Save the intended website origin on login items. Browser autofill requires the top page and target frame to match its scheme, host, and port exactly.
• Back up the whole data directory so the database and any managed instance-secret file stay together. Rotating security.encryptionSecret follows the same key-ring procedure as other encrypted company data.