What Ships Today

From threat detection to compliance — everything included. Here's what's available now and what's coming next.

Available Now

  • Multi-stage runtime security layer — every tool call authenticated, classified, and audited
  • Python SDK: clampd.openai(), clampd.anthropic(), LangChain, Google ADK, MCP proxy
  • TypeScript SDK: clampd.openai(), clampd.anthropic(), MCP proxy
  • Comprehensive built-in rules (SQL injection, SSRF, prompt injection, XSS, PII exfiltration, reverse shell, and more)
  • Scope-based permissions — least-privilege access with exemptable rules and constrained patterns
  • Policy engine with custom policy support for organization-specific rules
  • Risk scoring with behavioral baseline anomaly detection and auto-suspend
  • Full audit trail with PII masking and tokenization
  • Emergency kill switch (<25ms verified)
  • Dashboard UI — kill, suspend, or pause agents; manage policies, edit rules, live risk feed
  • Custom rules DSL with hot-reload
  • Cross-agent correlation for coordinated attack detection
  • CLI tool with live TUI dashboard and demo mode
  • Docker Compose deployment
  • Agent registry with scoped credentials
  • Circuit breakers, rate limiting, and graceful degradation
  • PyPI + npm packages published
  • Prompt and response scanning — PII detection, data anomaly scoring, poisoned response detection
  • A2A delegation — auto-detect agent chains, graph visualization, enforce approval policies
  • OIDC SSO (Keycloak, Okta, Azure AD)
  • Compliance export (SOC 2, GDPR)
  • Webhook notifications with event filtering

Coming Soon

  • Self-hosted marketplace (AWS, GCP, Azure)
  • Hybrid ML escalation — ambiguous risk scores auto-escalate to ML model for deeper classification
  • On-device intent classification for air-gapped deployments
  • Bring your own classification model (hosted endpoint or local)
  • LLM-as-judge — use an LLM to evaluate ambiguous tool calls in real-time