Security and trust
Current capabilities and open enterprise-readiness work. Items marked ⏳ are pending and are not available commitments.
Agent controls
RelayAuth supports scoped agent credentials. Cloud supports cancellation of individual workflow runs and revocation of individual tokens. Human approval gates can be configured per flow; approval is not a universal requirement for every write.
Cloud sign-in uses Google OAuth. Organization roles are owner and member. Workflow run history is available; it is not a comprehensive customer-visible security audit log.
⏳ Workspace-wide emergency controls, general read-only/suggest mode, enterprise SSO (SAML/OIDC), and a customer-visible security audit log.
Deployment and data flow
The hosted Cloud product connects authorized integrations, executes agent workflows in sandboxes, and records run data. Connected tools and model providers receive data required for the requested work. Cloud supports bring-your-own model keys.
The open-source core and Software Factory can run in your infrastructure. External integrations and model APIs still receive data when you configure them. The full Cloud product does not currently have self-host packaging.
⏳ Single-tenant and private-cloud packaging for the full Cloud product.
Data protection and retention
Cloud uses Cloudflare and AWS infrastructure and a Neon database. Confirm region and feature-specific data flows with us before a pilot.
Workspace run-data retention is configurable. Its default is to keep data indefinitely. Account deletion currently records a request for manual handling; automated completion is pending.
Review the privacy policy for collection, model processing, retention, and deletion details, and telemetry documentation for opt-out instructions.
⏳ A contractual no-training clause and confirmed zero-data-retention coverage for all model providers. We do not represent these as completed.
Provider inventory — unconfirmed
This is a preliminary inventory from repository evidence, not a finalized subprocessor register. Actual production use, processing locations, contracting entities, and downstream model providers require confirmation. Not every provider is used for every customer.
| Providers | Potential processing purpose | Verification |
|---|---|---|
| Cloudflare, AWS, Neon | Hosting, storage, database, networking | Locations and production scope unconfirmed |
| Daytona, E2B, Vercel Sandbox, Freestyle, Agent37, Microsandbox | Agent execution | Active providers and locations unconfirmed |
| Anthropic, OpenAI, xAI, Google Gemini, OpenRouter, Cursor | Model inference; OpenAI session embeddings | Routing, locations, and retention terms unconfirmed |
| Nango, Composio, GitHub, Slack, Dropbox | Authorized integrations and data movement | Feature-specific scope unconfirmed |
| Recall.ai, Modal, Hookdeck | Meeting recording, transcription, webhooks | Feature-specific scope and locations unconfirmed |
| Supermemory, Typesafe.ai | Agent memory and session evaluation | Production traffic unconfirmed |
| Stripe, PostHog, Resend | Billing, analytics, transactional email | Locations and contract details unconfirmed |
Compliance and security review
⏳ SOC 2 platform and auditor engagement, audit dates, and penetration-test attestation. No completed SOC 2 or penetration-test claim is made here.
For a pilot, agree the data boundary, integration scopes, approval gates, retention, and support terms with us before granting access.
Report vulnerabilities through our vulnerability disclosure policy.