Your data stays on your Mac.

OMEGA was built on a privacy-first architecture: local-first execution, your own API keys, signed safety policies, and an air-gap roadmap for Enterprise. The same patterns that make OMEGA fast and cheap also make it the easiest AI agent on the market to defend in a security review.

How we keep your data yours.

Local-first execution

OMEGA runs on your Mac. Your conversations, your memory, your Automations, and your data live on your hardware. Nothing routes through OMEGA's servers in normal operation. The only outbound traffic is the runtime you configured: your own API keys, a supported Codex or Claude subscription connection, or local-only MLX.

BYOK and subscription handling

API keys for OpenAI, Anthropic, Google, and other providers stay encrypted on your device. Codex and Claude subscription connections are handled as local runtime credentials. OMEGA never stores, proxies, or sees your raw tokens. We can't bill you per-token because we never see the requests.

Encryption at rest and in transit

Local credentials and application secrets are protected with macOS Keychain-derived keys. Workspace and runtime boundaries authenticate every internal service call; selected external providers receive only the data required for the user-authorized request.

Signed-and-pinned safety policies

OMEGA's safety policies are signed and hash-pinned. They can't be silently changed: not by us, not by an attacker, not by the model. Tampering shuts the system down before it runs.

Enterprise air-gap deployment

Zero-network operation, outbound allowlist enforcement, and local-model-only execution are on the Enterprise roadmap, scoped case-by-case on custom terms. Standard plans remain local-first; full air-gap controls will be Enterprise-only.

Audit logging

Every agent action (tool calls, model invocations, file accesses, costs) is recorded in an append-only Action Journal. The journal is yours, on your Mac, and surfaces in the in-app dashboard.

Encryption details, surface by surface.

What’s encrypted, how it’s encrypted, and where the keys live. No hand-waving.

SurfaceMethodNotes
API keys and subscription connectionsmacOS Keychain (AES-256, hardware-backed where available)Keys and subscription connection metadata live in your login keychain; OMEGA reads them only at the moment of an outbound provider call.
Local conversation + memory dataBundled Postgres on a Unix socket; file-system-level FileVault encryptionPostgres database in ~/Library/Application Support/OMEGA/, accessible only by your macOS user.
Internal pub/sub (Gateway → Agent)Bundled Redis 6380 with HKDF-derived password from per-install jwt_secretEven another process on your Mac can't read the bus without the secret.
License + account data on the licensing serverTLS 1.3 in transit, AES-256 at rest, signed JWTscentral.myomega.ai stores only email + license info: no conversation content, ever.
App binary + updatesApple Developer ID code signing, Apple Notarization, Tauri signed updaterHardened runtime + library validation. Updates are signature-verified before install.

What we’re defending against.

The threats we’ve thought through, and the controls that mitigate them. This is a living document; if you find a gap, please tell us.

Compromised AI provider keys

Mitigation: Keys live in your Keychain only; OMEGA can't leak what it doesn't see. Revocation is instantaneous via the provider's dashboard.

Malicious third-party MCP server

Mitigation: User-installed MCP servers are sandboxed and require explicit per-app approval gates before agents can call them. Always-allow lists are revocable from Settings.

Prompt injection / jailbreak attempts

Mitigation: Sensitive-action denial layer + Soul governance with hash-pinned policy. Risky tool calls trigger human-in-the-loop approval queue, not silent execution.

Runaway agent token costs

Mitigation: Neural-Fractal Agentic AI™ (NFA) budget authority enforces token + dollar caps at every cognitive scale. Per-task / per-day / per-week / per-month budgets in Settings.

Stolen device with active OMEGA install

Mitigation: FileVault encryption (recommended); license-server can deactivate stolen devices on request; Keychain protected by user password + device PIN.

Compromised licensing server

Mitigation: Server stores license metadata only: no conversation content. License JWTs are signed; offline grace period allows continued use even if the licensing server is unavailable.

Supply-chain attack on a dependency

Mitigation: SBOM generated per release; dependencies pinned by hash; reproducible builds; release artifacts signed by Apple Developer ID.

Where we stand.

We're honest about what's in place today versus what's in flight. Anything dated below is committed to in writing for Enterprise customers.

SOC 2 Type II (cloud surface scope)Planned
GDPR data exportAvailable today
CCPA / CPRA complianceAvailable today
DPA template (Pro / Team)On request
HIPAA controls (Enterprise)Roadmap
ISO 27001Planned · 2027
SSO / SAMLRoadmap

Independent reviews.

Reports are available under NDA via the legal contact form. Public summaries land here as audits complete.

Coming Q3 2026·Third-party penetration test
Scope
Native macOS app + licensing server
Auditor
TBD (NCC Group / Trail of Bits / Cure53 shortlist)
Deliverable
Executive summary public; full report under NDA
Ongoing·Internal red-team review
Scope
Each release before signing
Auditor
Internal security review
Deliverable
Findings closed before release; high-severity items disclosed publicly

Found a vulnerability?

We welcome reports from security researchers. Our policy is below; operating within it protects you legally and earns you public credit (if you want it).

How to report

  1. 1. Submit through the security contact form
  2. 2. Include: affected component, reproduction steps, impact, and any proof of concept
  3. 3. We acknowledge within 24 hours and triage within 2 business days

Our commitments

  • We acknowledge security reports within 24 hours.
  • We triage and assign severity within 2 business days.
  • We commit to a fix timeline based on severity (Critical: 7 days; High: 30 days; Medium/Low: next release cycle).
  • We coordinate public disclosure with the researcher once a fix is available.
  • We provide public credit (with permission) on this page's Acknowledgments section.
  • We do not pursue legal action against good-faith security researchers operating within our policy.
Safe Harbor: Good-faith security research conducted in accordance with this policy will not result in legal action against you. We support the disclose.io core terms.

What’s in scope

In scope: OMEGA native macOS app, central.myomega.ai, and the myomega.ai marketing site.

Out of scope: third-party AI providers (report to them directly), social engineering, physical attacks, denial of service, automated scanners against production, and pre-release features explicitly marked unstable.

Bug bounty: not currently offered; we credit researchers publicly and provide swag for substantive findings. A formal bounty program is on the 2027 roadmap.

Hall of fame.

Researchers who responsibly disclosed security issues, with their permission to be credited publicly.

None yet. Be the first. Report a vulnerability →

Every vendor that touches your data.

Full transparency on every third party we use that ever sees account or billing data. We’ll notify you 30 days before adding any new sub-processor that touches customer data.

Stripe, Inc.

Payment processing, billing, subscription management

Privacy Policy
Data accessed
Email, billing address, payment method
Location
United States (multi-region)
Certifications
PCI DSS Level 1, SOC 1, SOC 2

Resend, Inc.

Transactional email delivery (account activation, receipts, support)

Privacy Policy
Data accessed
Email address, name
Location
United States
Certifications
SOC 2 Type II

Twilio Inc.

Phone verification at signup, where required (Verify one-time SMS codes + Lookup line-type checks)

Privacy Policy
Data accessed
Phone number, verification metadata
Location
United States
Certifications
SOC 2 Type II, ISO 27001

Vercel, Inc.

Marketing website hosting + edge CDN

Privacy Policy
Data accessed
Public website traffic logs (anonymized IPs)
Location
Global edge network
Certifications
SOC 2 Type II, ISO 27001

Cloudflare, Inc.

DNS, DDoS protection

Privacy Policy
Data accessed
Public website traffic (no payload inspection)
Location
Global edge network
Certifications
SOC 2 Type II, ISO 27001, PCI DSS

Hostinger International Ltd.

Licensing server VPS hosting

Privacy Policy
Data accessed
Encrypted database snapshots, server logs
Location
United States, Lithuania
Certifications
ISO 27001
Subscribe to sub-processor change notifications: Pro and Team customers can subscribe to email notifications via the legal contact form (subject: “Sub-processor notifications”).

Need more for procurement?

Sub-processor list and DPA template are above. For penetration test reports, security questionnaires, or custom procurement requirements, contact us.