Security
Every extension is scanned.
No exceptions.
Our 4-layer antivirus pipeline analyzes every submission before it reaches you. Extensions run in isolated WASM sandboxes with capability-based permissions.
Pipeline
4-layer security pipeline
Each layer produces a score from 0-100. The weighted average determines the verdict. No single point of failure.
Signature Scanning
30%Every WASM binary is checked against 79+ known malicious patterns. Our signature database covers command injection, data exfiltration, cryptomining payloads, reverse shells, and more. Patterns are compiled into a RegexSet for fast matching.
Heuristic Analysis
25%Behavioral rules analyze whether an extension's permission requests make sense for its stated purpose. A weather tool requesting filesystem write access or a calculator asking for network permissions will be flagged.
AI Code Review
30%Claude analyzes the extension's manifest, permission list, and embedded code patterns. The AI evaluates data exfiltration risk, checks whether behavior matches the description, and assesses whether permission requests are justified.
Sandbox Testing
15%The extension is executed in a fully isolated WASM runtime with synthetic inputs. We monitor memory usage, CPU time, attempted syscalls, and network call attempts. This catches runtime threats that static analysis might miss.
Isolation
WASM Sandboxing
Every extension runs inside an isolated WebAssembly sandbox powered by Wasmtime. Extensions cannot access your system directly — they can only use host functions that you've explicitly granted through the permission system.
The sandbox enforces strict resource limits: maximum memory allocation, CPU time per tool call, and concurrent execution limits. If an extension exceeds its limits, it's terminated immediately without affecting other extensions or the Omni runtime.
Cannot access
Raw filesystem
Network sockets
System processes
Other extensions' data
Clipboard (without permission)
Environment variables
With permission
HTTP to allowed domains
Scoped file read/write
LLM inference via bridge
Channel messaging
Extension key-value store
Permissions
Capability-based permissions
Extensions declare what they need in their manifest. You review and approve permissions before installation. Every capability has a severity level.
network.httpmediumnetwork.websocketmediumfilesystem.readhighfilesystem.writehighai.inferencemediumchannel.sendhighbrowser.scrapehighclipboard.readmediumclipboard.writelowsystem.notificationslowsystem.schedulinglowstorage.persistentlowMonitoring
Continuous monitoring
Security doesn't stop at initial publication. We continuously monitor the extension ecosystem to catch threats that emerge after initial review.
Weekly re-scans
All published extensions are periodically re-scanned against updated signature databases to catch newly discovered threats.
Updated signatures
Our signature database is regularly updated with new patterns discovered by the security community and our own research.
Publisher notifications
If a re-scan detects issues in a previously clean extension, the publisher is notified immediately and given time to address findings.
Community reporting
Users can report suspicious extensions through the marketplace. Reports trigger additional scans and human review.
Trust
Trust levels
Extensions are classified based on their publisher's track record and verification status.
Verified
Highest trust
Publisher identity verified. Source code reviewed. Manually audited by the Omni security team.
Community
Standard
Published by an established member with positive track record. Passed automated scanning. Most extensions fall here.
Unverified
New
New publisher or extension without track record. Passed automated scanning but not manually reviewed. Review permissions carefully.
Disclosure
Responsible disclosure
We take security vulnerabilities seriously. If you discover a security issue in the Omni platform, marketplace, or any published extension, please report it responsibly.
We acknowledge valid reports within 48 hours, provide regular status updates, and do not pursue legal action against researchers who act in good faith.
Email security@omniapp.org with a detailed description
Include steps to reproduce, affected versions, and impact
Allow 90 days to address the issue before public disclosure
Do not exploit beyond what is necessary for demonstration