Open source core. Enterprise platform.

Start free. Upgrade when your team needs the dashboard, the brain, and the compliance reports.

Open Source
Free forever
For individual developers protecting their agents.
  • Detection layers 1-3 (structural, anomaly, CFI)
  • Taint propagation (6-level data trust)
  • Tool execution order enforcement
  • Agent mandate (per-session scope)
  • Memory poisoning defense
  • Output sanitization
  • Policy engine (default-deny)
  • MCP proxy (Claude Code / Cursor)
  • CLI + JSON audit log
Team
$49/agent/mo
For teams that need visibility and the full detection engine.
  • Everything in Open Source
  • Full 7-layer detection (F1 = 0.94)
  • Real-time dashboard (5 views)
  • Agent communication bus
  • Governed memory (privilege-tiered brain)
  • SIEM export (Splunk, Datadog, webhook)
  • Pre-decision safety gates
  • Business impact scoring
  • Cloud API (remote agents connect)
Enterprise
Custom pricing
For organizations with compliance, federation, and self-hosting needs.
  • Everything in Team
  • Self-hosted (Docker Compose, Helm, air-gapped)
  • OIDC / SCIM federation (Okta, Azure AD)
  • Multi-tenancy (BU isolation)
  • Compliance reports (SOC 2, PCI DSS, HIPAA)
  • Human-in-the-loop approval workflows
  • Auto-quarantine (self-defending agents)
  • Pre-deployment security scanner
  • Ticketing integration (ServiceNow, Jira)
  • Dedicated support + SLA

Common questions

What's included in the free tier?

Yes. It catches prompt injection, enforces tool scope, tracks data taint, validates tool ordering, and produces a JSON audit log. Developers use it in production today. The paid tiers add the dashboard, full 7-layer detection, governed memory, and enterprise features.

What's the latency impact?

P50 is 0.39ms for the full 7-layer pipeline. The OSS layers (1-3) run in under 0.05ms. Tool calls to LLMs take 500ms-5s — Agentic Glass overhead is invisible.

Can I self-host?

Enterprise tier includes Docker Compose and Helm chart for self-hosted deployment, including air-gapped environments. The platform runs entirely on your infrastructure — no data leaves your network.

What frameworks do you support?

LangChain, CrewAI, AutoGen, OpenAI Agents SDK, Anthropic, Claude MCP, and any Python agent framework. The SDK auto-detects and patches at import time. The MCP proxy works with Claude Code, Cursor, and any MCP client.

Estimate your tier

15 agents100
Recommended tier
Open Source
Free forever

Feature comparison

FeatureOSSTeamEnterprise