Projects

Rampart intercepts agent tool calls at the OS layer before they execute. Network requests, file writes, subprocess spawns — all evaluated against your policy profile. If an action violates policy, it's blocked. The agent sees a failure; you see a log entry.

• — OS-layer interception via LD_PRELOAD and process shims
• — Policy-as-code: YAML rules, version-controlled
• — Works with Claude Code, Codex, Cursor, Cline, and any MCP server
• — Ships with default profiles tuned for common agents

Snare plants fake credentials in the locations agents look — AWS config, GCP service accounts, SSH config, dotenv files. When a compromised agent uses one, the embedded callback URL fires and you get an alert with the exact SDK, IP, and user agent that triggered it.

• — awsproc canary fires before the first network request — at credential resolution time
• — Alerts include SDK user agent, IP, ASN, and location
• — Detects both compromised agents and human attackers
• — 18 canary types: AWS, GCP, Azure, SSH, k8s, Hugging Face, and more