Product guides

Privacy & security

AverQel applies authentication, tenant and user authorization, and integration policies in its backend. These controls reduce unauthorized access; they do not mean that all content is client-only or invisible to the service.

Tenant and user isolation

Protected API routes scope reads and writes to the authenticated account, tenant, resource ownership, and role permissions.

Credential handling

Provider and OAuth credential material is handled by backend services and is not intended to be returned in normal frontend responses. Exact protection depends on configured encryption keys and deployment practices.

Tool authorization

Connected tools are governed by ownership, enabled status, provider policy, tool modes, risk limits, and required approvals.

Content boundaries

Documents and AI requests are processed by configured server-side services. Assigned Support Managers can read support tickets; assigned Feedback Managers can read submitted feedback text.

MCP account and tenant isolation

MCP connection lookups are scoped by tenant, user, and provider connection. DeepSpace tool execution checks the owning account and relevant connection and conversation permissions; frontend state is not authorization evidence. Review a provider's requested scopes and policy before connecting it.

OAuth consent and secret lifecycle

  • Users consent on the provider's authorization page; AverQel does not need the provider password.
  • Access tokens, refresh tokens, and client secrets are handled server-side and encrypted at rest where the configured credential storage applies.
  • Only safe account details and granted scope information needed by the interface should be exposed to the owning user.
  • Disconnect removes local connection credentials and requests provider revocation where supported; remote revocation behavior is provider-dependent.

Permission modes and precedence

MCP controls can include Read-only, Always allow, Needs approval, or Blocked modes, along with connection, conversation, and risk limits. A more restrictive applicable rule should prevent a tool call. Always allow does not bypass backend authorization, provider state, or platform safety rules. Review the connector guide for current precedence details.

AI providers and content processing

When a user sends a prompt or document question to an external model provider, the backend sends the content needed for that request to the configured provider. Provider retention, processing location, and contractual terms are separate from AverQel's tenant access controls. Use a provider approved for the data you intend to process.

Qel Connect and encryption

Qel Connect chat includes client encryption helpers and an optional server-mediated sealed storage mode. The Qel Connect connection code is available through the API, and the sealed mode can be opened by the API for authorized members. Do not treat Qel Connect chat as zero-knowledge or server-blind end-to-end encryption. Qel Connect documents are processed server-side. See the Qel Connect guide for details.

Admin and operational boundaries

Administrative access is role- and route-specific, and platform-wide routes require an account in the configured platform-admin allowlist. Support content access requires the separate Support Manager role; these managers can read support tickets and attachments across workspaces and must also be platform-admin allowlisted. Feedback access is separate and requires the assigned Feedback Manager role; it can read feedback text across workspaces, but its staff API omits submitter email and account IDs. Designated Qel Connect moderators can read report details only inside their active tenant. These are content-bearing workflows, so do not treat AverQel as a system where administrators can never access user content. Privileged support, feedback, and moderation reads and changes are recorded in the audit log.

The default account view shows email, status, roles, and MFA status without per-user usage counts or recent activity. Workspace analytics use a rolling 30-day window with a five-member minimum group threshold; smaller groups and daily points are hidden. Global worker metrics require platform-admin access, and the public gateway blocks the raw Prometheus route.

Operators must protect signing and encryption keys, database and object-storage access, OAuth credentials, backups, logs, network egress, and administrative accounts. Production security depends on deployment configuration, patching, monitoring, incident response, and verified restore procedures as well as application code.