Identity and least privilege
Users, service tokens, repositories, properties and CMS connections are authorized separately and should receive only required scopes.
Explore SEOForge
Security model
Agents receive bounded tools—not credentials or inherited production authority.
SEOForge separates identity, provider access, agent execution and production approval. This page describes design controls, not a claim of invulnerability or an unearned certification.
Nothing publishes without approvalPaid product accessNo ranking guarantees
Production authority remains locked
- Explicit tenant and site context
- Credentials withheld from agent prompts
- Human approval for exact changes
- Revocable provider connections
Control layers
Reduce what each component can see and do.
Secret handling
Provider credentials are encrypted or hashed as appropriate, redacted from agent context and never available through API, CLI or MCP retrieval.
Tenant and tool boundaries
Every provider observation, run and artifact carries workspace and site scope; tools are allowlisted for the assigned role.
Detection and recovery
Audit events, provider health, retry state and production validation support investigation. No system is completely secure.
Prompt-injection posture
Treat websites, repositories and provider content as untrusted input.
- Separate retrieved content from system policy
- Allowlist tools by role and assignment
- Do not expose secrets in model context
- Validate targets and arguments server-side
- Require human approval for production effects
- Record source and concise decision evidence
Certification status
This page does not claim SOC 2, ISO 27001, PCI, HIPAA or any other certification. Completed attestations will be named only when current supporting documentation is available.
Paid deployment · human authority
Review the actual data flow and permissions for your deployment.
Contact sales for a security conversation and available evaluation documents.