Glossary
Overview
Definitions of the terms used across this documentation. Cerbera AI is built to serve technical and non-technical stakeholders alike, so this page keeps each definition short and practical.
Core Terms
| Term | Definition |
|---|---|
| Proxy | A program that sits in the network path and inspects traffic as it passes. Cerbera AI installs a lightweight local proxy on each device. See How It Works. |
| MITM (man-in-the-middle) | A position in the request path that can read and modify traffic. A security proxy uses this technique deliberately to inspect AI traffic. |
| TLS interception (HTTPS offloading) | Decrypting and re-encrypting HTTPS traffic so its contents can be inspected. It requires a trusted certificate on the device. |
| Leaf certificate | A per-device certificate used for TLS interception. Cerbera generates one on each device and holds no root CA, so a breach of Cerbera does not yield a key to intercept your traffic. |
| Root CA (certificate authority) | The top-level certificate that can issue others. Cerbera deliberately does not hold one. |
| MDM (mobile device management) | The system used to manage company devices (for example Jamf or Intune). Cerbera AI deploys through it in one click. |
AI & Agents
| Term | Definition |
|---|---|
| LLM (large language model) | The model behind tools like ChatGPT, Claude, and Gemini. |
| Agent | An AI tool that takes actions on a device, such as Claude Code running commands or editing files, not just answering questions. |
| MCP (Model Context Protocol) | An open protocol that lets AI clients call external tools and APIs. See MCP. |
| MCP server | A service an AI client connects to over MCP. Can be official (provider-maintained) or third-party (community-maintained). |
| Shadow AI | AI tools used inside an organization without security review or approval. |
| Prompt injection | An attack that hides instructions in content an AI reads, hijacking its behavior. Detection is on the Cerbera roadmap. |
Rules & Actions
| Term | Definition |
|---|---|
| Monitor | A rule action that logs a match without changing or blocking anything. |
| Redact | A rule action that redacts secrets (API keys, AWS keys, PII) from a request before it leaves the device. See Rules. |
| Block | A rule action that stops a request and shows the user a pop-up explaining why. |
| Allow-by-default | Unrecognized traffic is permitted. Cerbera AI enforces this for traffic it does not identify as a known AI tool, so an unexpected provider API change does not break users. |
| Deny-by-default | Anything not explicitly allowed is blocked. Recommended only after monitoring has surfaced the legitimate tools in use. |
Adjacent Tools
| Term | Definition |
|---|---|
| EDR (endpoint detection and response) | Watches system calls on a device. Complementary to Cerbera AI but cannot read prompts or govern AI network calls. |
| Firewall / SWG (secure web gateway) | Inspects network traffic. Not designed to understand AI-specific behavior. |
| ZTNA (zero trust network access) | Controls access to internal resources (for example Cato Networks). Chains with Cerbera AI rather than conflicting with it. |
| VPN | Routes traffic through a private network. Cerbera AI behaves similarly in that traffic passes through it, but its purpose is AI inspection. |
Data & Integrations
| Term | Definition |
|---|---|
| OpenTelemetry (OTel) | An open standard for telemetry. Cerbera AI exports alerts and rule matches in this format. See Openness & Interoperability. |
| SIEM | A security information and event management system that aggregates logs and alerts. Cerbera AI feeds it over OpenTelemetry. |
| Webhook | An HTTP callback that pushes events to another system as they happen. On the Cerbera roadmap. |
| IaC (infrastructure as code) | Managing configuration declaratively in version control rather than by hand. On the Cerbera roadmap. |
| P75 / P99 latency | The latency experienced by the 75th and 99th percentile of requests. Cerbera AI runs under 1 ms at P75 and around 10 ms at P99. |