Title here
Summary here
Security
This page is intentionally concise and stable. It summarizes the security model for architecture, DSI/RSSI review, and run operations.
For exact implementation details and fast-changing configuration keys, use the linked GitHub sources at the end.
Erasure you can prove — How Fred deletes a conversation completely and provably, across every store, on the team’s own terms (RGPD-ready data erasure, by design): Erasure you can prove.
Security Scope
Fred security is organized around five domains:
- identity and access
- service-to-service trust
- runtime isolation
- data and secrets protection
- auditability and operational control
Identity And Access
End-user authentication
- User authentication is based on OIDC/OAuth2 (typically Keycloak).
- UI obtains a user token and calls backend APIs with bearer authentication.
- Backends validate token signature and standard claims.
Authorization model
- RBAC is enforced on protected API surfaces.
- ReBAC policies can be enabled for team and resource-scoped access.
- Admin-only capabilities must stay restricted to administrator roles.
- Detailed role and permission matrix is documented in Access Model.
Service-To-Service Security
- Agentic and Knowledge Flow communicate with authenticated service identities.
- Client-credentials are used for service tokens where required.
- MCP integrations use authenticated outbound calls; transport/auth behavior is policy-controlled.
Runtime And Data Boundaries
- User session boundaries are enforced at runtime.
- Long-running processing uses Temporal workers with explicit task boundaries and retries.
- Storage architecture is deployment-dependent (PostgreSQL and optional ClickHouse).
- Data governance decisions (models/tools/prompts/agents/data scopes) should be explicit at team level.
Secrets, Keys, And Transport
- Secrets are managed outside source code (K8s secrets, vaults, CI/CD secrets).
- TLS is required for external endpoints and strongly recommended end-to-end.
- Token/key rotation procedures are part of platform runbooks.
- Logs and traces must avoid leaking credentials or full tokens.
Operational Security Baseline
Before production go-live, teams should validate:
- IdP configuration (realm, clients, roles/groups, token claims).
- RBAC/ReBAC enforcement for admin actions.
- Secret distribution and rotation process.
- TLS and ingress hardening.
- Audit log retention and incident response workflow.
- Dependency and image vulnerability scanning in CI/CD.
Governance Position
Fred is policy-first for enterprise usage:
- policies are the authority for model/tool/prompt/agent/data access decisions,
- runtime behavior is expected to be deterministic and auditable,
- end-user convenience must not bypass governance controls.
Related References
- Erasure you can prove — RGPD-ready data erasure, by design
- Architecture
- Deploying on GCP / GKE
- Access Model
- Policy-based LLM Routing