Security & Trust

Your org's data never has to leave your control

SF Agent works on metadata, not data. The platform reads your org's blueprint — object definitions, flows, Apex, permission sets — to plan and build changes. It does not read, store, or transmit your customer records, and it never needs to.

Production deploys are yours alone. SF Agent builds and tests in sandboxes, then hands you a deployment package that your production org validates via Salesforce's own checkOnly mechanism — nothing deployed, tests run, org says "this would succeed." Promotion happens only when you click deploy, and only for the exact package your org already validated. Every step is recorded in a tamper-evident audit trail.

Data sovereignty — choose your guarantee

Four tiers, from strong defaults to nothing-leaves-your-perimeter. Every tier keeps customer records out of scope entirely.

1

Standard

Metadata only, redacted, retained on your terms
  • SF Agent reads your org's configuration metadata — fields, flows, classes — never your customer records or CRM data.
  • PII-pattern redaction runs before anything reaches a model provider; egress policy is enforced per tenant (none / summary-only / approval-gated).
  • Secrets are AES-256-GCM encrypted, write-only after storage. Run artifacts follow your retention policy.
2

Your model contract

AI traffic runs under YOUR provider agreement
  • Bring your own OpenAI / Anthropic / Azure key: every model call executes under your enterprise agreement — your DPA, your zero-data-retention terms.
  • SF Agent is contractually out of the model-data path. Per-tenant provider allowlists prevent unapproved model usage.
3

Your cloud for models

Model calls never leave your cloud account
  • Route model traffic through Claude on AWS Bedrock, Vertex AI, or Azure OpenAI inside your own cloud tenancy.
  • Your metadata travels from the platform to your cloud and back — the model vendor's consumer surface never sees it.
4

Single-tenant / self-hosted

Nothing leaves your perimeter, period
  • The entire platform is a two-piece stack (Next.js + Postgres) deployable inside your VPC.
  • Combined with in-tenancy model routing, no metadata, artifact, or log ever exits your cloud account.

Platform security controls

• Multi-tenant isolation on every query
• AES-256-GCM secrets, write-only after storage
• Role-based access on all sensitive routes
• Append-only, hash-chained audit trail
• Outbound endpoint validation + allowlists
• PII redaction before any model egress
• Per-tenant AI egress policy enforcement
• DEV-only autonomous deploys by policy
• HttpOnly secure session cookies
• Login brute-force lockout + MFA support
• Per-call token & cost accounting
• Deployment consent + approval gates

Operational questions, vendor security reviews, and incident reports: security@sfagent.dev.

Additional documents: Subprocessors, Data Processing Addendum, Vulnerability disclosure.

Back to home
Security & Data Sovereignty | SF Agent | SF Agent