Trust center
Security posture for reviewers
Runline is AI runtime enforcement on the endpoint, not an LLM proxy. This page summarizes how we isolate customer data and operate the control plane. Detailed architecture lives in our operator documentation; request the full pack during a POV.
Threat model
Runline enforces at AI-driven CLI and MCP execution on enrolled endpoints. In scope: destructive commands from Cursor, Claude Code, and similar agents. Out of scope (v1): browser-only AI sessions and human Terminal use on the same machine.
Tenant isolation
POV and Team customers receive a dedicated tenant with isolated PostgreSQL (DB-per-tenant on RDS). Row-level security enforces tenant_id on every data-plane query. SSO email domains are mapped before first login, unmapped domains are rejected in strict mode.
Control plane
Console and agent APIs run on AWS (Fargate + RDS). Agents enroll with one-time install tokens, then receive per-endpoint bearer credentials and mTLS client certificates. Audit metadata is stored in your tenant partition, not full command output by default.
Compliance
- SOC 2: in progress, evidence package available on request during enterprise evaluation.
- Data residency: POV tenants run on Runline-operated AWS (US). Dedicated stacks and region pinning available for Enterprise.
- Subprocessors: AWS (compute, database), Cloudflare (edge), Resend (contact email), full list in enterprise DPA.
Honest v1 limits
PATH bypass is possible until Apple Endpoint Security ships. Reference platforms are macOS and Linux; Windows enforcement is in preview. Slack interactive approve is planned; webhooks and console grants today. We disclose these in every POV scoping call.