Menu
Blog How it works Use Cases
agentsh
Open-source runtime for AI agent security
Beacon
AI endpoint visibility and control
Watchtower
Enterprise control plane for agentsh and Beacon
Request Access
Command and Control

Govern AI execution,
everywhere.

Watchtower is the command center for Beacon and AgentSH. Centralized policies, approvals routing, SIEM forwarding, and a fleet-wide kill switch.

  • Central policy Write once. Target by endpoint group, team, tool, workload type, or environment.
  • Approvals routing Slack, email, SMS. Full context. Durable decisions.
  • Evidence and compliance Retention, audit trails, and SIEM export.
  • Emergency brake Pause AI execution fleet-wide in seconds.
watchtower
agents: 847 connected
policies: 12 active
events/hr: 23,491
uptime: 14d 7h 23m
Last incident: 3d ago (agent-dev-17 blocked)
watchtower
approvals:
production-deploy:
channels: [slack:#security-ops]
sensitive-files:
channels: [email:security@acme.com, sms:oncall]
default:
channels: [slack:#agent-alerts]
Approval routing configured
3 rules active across 847 agents
watchtower
⚠ EMERGENCY KILL SWITCH ACTIVATED
Pausing all agent operations...
847 agents paused
0 pending actions cancelled
All new requests blocked
Fleet halted in 1.2s
Run 'watchtower resume' to restore operations
Control plane

One control plane for supervised and unsupervised AI.

Beacon secures supervised AI on endpoints. AgentSH secures unsupervised agents in CI, containers, and dev environments. Watchtower governs both from one place.

Supervised

Beacon on endpoints

Desktop AI tools — Claude, ChatGPT, Cursor, Claude Code — monitored and controlled on macOS and Windows with human-in-the-loop approvals.

Learn about Beacon →
Unsupervised

AgentSH on runtimes

Headless agents in CI pipelines, containers, sandbox runners, and dev environments — policy enforced at the syscall level with no UI required.

Learn about AgentSH →
Policy

Target policy to how AI runs in your org.

By tool

Per-application rules

Allow Claude to reach Anthropic APIs. Block Cursor from unknown registries. Different tools, different policies.

By environment

CI, staging, production

Tight lockdown in production. Broader access in dev. Prompt for deploy actions in CI pipelines.

By sensitivity

Secrets and credentials

Prompt before reading secrets paths. Redirect package registries to internal mirrors. Block exfiltration patterns.

Approvals

Approvals that work at scale.

# watchtower.yaml — approval routing

approvals:
production-deploy:
channels: [slack:#security-ops]
timeout: 300s

sensitive-files:
channels: [email:security@acme.com, sms:oncall]
timeout: 60s

default:
channels: [slack:#agent-alerts]
cache: true

Route to the right channel

Send approval requests to the right on-call team based on policy — Slack, email, or SMS.

Full context in every request

Tool, process chain, destination, command, and scope. Reviewers see exactly what is being requested.

Cache decisions

When appropriate, cache approvals so teams do not get spammed with the same request.

Slack Email SMS Webhook
Evidence

Forward everything to your security stack.

Export events and decisions

Splunk, Sentinel, QRadar. Your existing security tools, your existing workflows.

Raw evidence for investigation

Keep the full record for audits and incident response.

AI execution in one place

Make AI execution visible in the same pane as the rest of security operations.

Splunk Microsoft Sentinel QRadar OpenTelemetry
Kill switch

When things go wrong, stop the bleeding.

Fleet Halted
All endpoints and agent runtimes paused

One action, fleet-wide

Pause AI execution across every endpoint and agent runtime in seconds.

Resume when ready

Investigate the incident, update policy if needed, then restore operations.

Scoped or global

Halt everything, or target specific teams, environments, or tool types.

Plans

Scale as you grow.

Team

For small teams
  • Centralized policy management
  • Approvals routing (Slack, email)
  • Fleet dashboard
  • 7-day event retention
  • Community support

Enterprise

For organizations at scale
  • Everything in Team
  • Multi-policy groups
  • 30/90/365-day retention
  • SIEM integration (Splunk, Sentinel, QRadar)
  • SSO and role-based access
  • Fleet-wide kill switch
  • Dedicated support
See use case: Incident response Learn about Execution-Layer Security
FAQ

Common questions.

What's the difference between agentsh, Beacon, and Watchtower?

agentsh is the open-source runtime that enforces policies on AI agents. Beacon monitors AI desktop tools on your endpoints. Watchtower is the enterprise control plane for both — centralized management, fleet-wide visibility, and compliance integrations.

How does deployment work?

agentsh runs alongside your AI agents (CI runners, containers, dev environments). Beacon installs on endpoints where AI tools run (laptops, workstations). Watchtower connects to both and provides a single pane of glass for policy management and monitoring.

What platforms are supported?

Linux (full enforcement via eBPF/LSM), macOS (ESF or FUSE-T), and Windows (minifilter driver). Coverage varies by platform.

Can I start with open source and upgrade later?

Yes. Start with agentsh or Beacon on your team's machines, then add Watchtower when you need centralized control or compliance features.

What compliance standards does this help with?

Watchtower provides audit trails and SIEM integration that support SOC 2, ISO 27001, and similar frameworks. We can discuss your specific requirements.

Talk to us.

Tell us about your environment and we'll reach out to discuss how Watchtower fits your security needs.

Whether you're exploring AI security for the first time or looking to scale existing deployments, we'll help you find the right approach — from open-source agentsh and Beacon to full enterprise control with Watchtower.