Skip to content

Security and data handling

Your instance serves your organization only. It has its own application, its own database and its own dedicated model capacity, kept separate from every other Hacker Sidekick customer and from the public Hacker Sidekick service.

  • Members only. Every request, from Desktop or the admin portal, is checked against your organization’s membership. A valid Hacker Sidekick account that isn’t a member is refused, and a sign-in that’s valid for another customer’s instance doesn’t work on yours.
  • Fails closed. If membership can’t be confirmed (for example, the Hacker Sidekick sign-in service is briefly unreachable), the request is refused with a “try again shortly” error. It’s never let through by default.
  • Fast revocation. Membership is re-confirmed at least every 5 minutes, so removing someone takes effect within about 5 minutes, not when their sign-in eventually expires.
  • Your sign-in protections apply. Members sign in through their Hacker Sidekick account, so the MFA or passkey they’ve set up there protects the instance too.
  • Server-side permissions. Admin-only actions are checked on the server for every request, and only the organization admin role counts as Admin. A standard member’s session never receives admin-only data: the member directory, Top users and the change log aren’t even queried for them.
  • Protected admin forms. Every admin form carries a per-session token, so another website can’t trick an admin’s browser into changing your configuration.
  • Confirmation before disruptive changes. Deletes, turning off pushed content, and changing who may add their own MCP servers all ask the admin to confirm first.

The MCP server credentials you enter in the admin portal are:

  • Encrypted before they’re stored, with a key held only by your instance, not in the database. A copy of the database alone doesn’t reveal them.
  • Never shown in the portal. MCP servers show only configured or none.
  • Never sent to members’ machines. MCP calls go through your instance, which adds the credential on the way out. See MCP Servers.
  • All traffic to your instance uses HTTPS (TLS).
  • The instance’s database runs on a managed Azure database service with disk encryption and automated backups.
  • Model requests go from your instance to its dedicated model deployments, not to shared capacity.
Record Contents Not recorded
Usage log Each model request: member, model, tier, token counts, response time, success or failure (with the provider’s status code) Prompt and response text, and the provider’s error message text
Change log Each admin configuration change, and each MCP tool call by tool name Credentials, prompt/template text, tool arguments and results
Member directory Email, role, first and last seen Anything else about the person

Usage records are kept for 400 days and change-log records for 2 years, then deleted automatically. Ask Hacker Sidekick if you need different periods.

Each account can start up to 600 requests a minute and have up to 30 in flight at once. These stop a runaway script or a leaked sign-in from exhausting your model quota, and are well above what normal Desktop use, including workspace indexing, needs.

  • Rotate MCP credentials in one place. When you rotate a credential at the source, paste the new value in the portal. Members don’t need to do anything.
  • Prefer org MCP servers to shared keys. Put internal tools behind an MCP server rather than pasting keys into skills or prompts.
  • Remember deletes are immediate. There’s a confirmation but no undo and no second approver.
  • Review the change log regularly, especially after adding admins.
  • Keep secrets out of text fields. Skills, prompts, names and descriptions are readable by every member.