AgentSoap processes outbound agent payloads to identify and redact sensitive information. We do not store the full content of sanitized payloads unless you have explicitly enabled "Payload Logging" for debugging purposes. Even then, data is encrypted at rest using industry-standard protocols.
Our semantic firewall uses non-persistent vector analysis to evaluate intent. Your agent's internal thought processes are never used to train global models; all analysis is performed within your isolated tenant environment.
We collect minimal metadata (API keys, timestamps, and risk scores) to provide telemetry and billing services. This data is never sold to third parties and is used exclusively for service maintenance and security auditing.
When using the AgentSoap MCP server with third-party platforms (Claude, OpenAI), metadata related to security blocks may be transmitted to those platforms if you have configured such integrations. Please review their respective privacy policies.