Security Policy

Reporting a vulnerability

If you discover a security vulnerability in Omnodex, please report it responsibly. Do not open a public GitHub issue for security vulnerabilities.

Email security@omnodex.com with a description of the vulnerability and its potential impact, steps to reproduce the issue, affected versions if known, and any suggested fix you may have.

We will acknowledge your report within 5 business days and work with you to understand and address the issue. We aim to provide a fix or mitigation plan within 30 days of confirmation, depending on complexity.

What qualifies

We are interested in vulnerabilities in the Omnodex software itself, including but not limited to:

• Event log tampering or integrity bypass
• Rule engine bypass (risk events that should fire but don't)
• Credential or sensitive data leakage through Omnodex's own operation
• Interceptor vulnerabilities that could affect the host agent's execution
• Encryption weaknesses in the sync encryptor or key derivation
• Authentication or authorization issues in cloud API interactions

Out of scope

• Vulnerabilities in upstream dependencies (report those to the dependency maintainer; let us know if Omnodex is affected so we can update)
• Issues in AI agents that Omnodex monitors (those are the agent platform's responsibility)
• Social engineering or phishing attacks against Omnodex team members
• Denial of service via resource exhaustion against local CLI tools

Coordinated disclosure

We follow coordinated disclosure. We ask that you give us reasonable time to address the issue before making it public. We will credit reporters in the fix announcement unless you prefer to remain anonymous.

Supported versions

During the current pre-1.0 development phase, security fixes are applied to the latest release only. We do not backport fixes to older versions at this time.

Security by design

Omnodex is built with several security principles at its core:

Zero token overhead

Interceptors run out-of-band from the agent's context window and never inject content into the agent session. Omnodex cannot alter the agent's behavior or consume your tokens.

Async and non-blocking

Interceptors append to the event log and exit. They never block the agent's execution path, so a bug in Omnodex cannot stall the monitored agent.

Append-only event log

The JSONL event log is the source of truth. Once written, events are not modified. The SQLite read model is a derived projection that can be rebuilt from the log at any time, ensuring auditability and tamper-evidence.

Zero-knowledge sync encryption

Cloud sync uses AES-256-GCM with client-side key derivation via Argon2id. The server never holds plaintext or key material. Your data is encrypted before it leaves your machine.

Privacy-preserving telemetry

Only statistical aggregates cross the wire: counts, timing distributions, hashed identifiers, and locally-computed risk scores. No raw tool names, file paths, credentials, or conversation content leave the machine.

Contact

For security-related questions or concerns, contact us at: