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
Use Cases

Where Execution-Layer Security matters most.

Start with the environments where AI execution is already happening: endpoints, CI, internal automation, and sensitive systems. Apply least privilege at runtime, and keep full audit trails.

Starting points

Choose your starting point.

1) Supervised copilots on endpoints

Beacon

Problem

Desktop AI tools can read files, access browser sessions, and connect to arbitrary services using user credentials. Supervised workflows still produce unsafe execution, especially when copilots automate actions.

What to enforce

  • Block unknown outbound destinations and new MCP servers by default
  • Prompt before accessing sensitive paths, clipboard, keychains, or browser automation
  • Allow only approved dependency registries, APIs, and internal services
  • Maintain evidence-grade audit logs for security review and incident response

Outcomes

  • Visibility into "shadow AI" usage across devices
  • Reduced exfiltration risk without banning tools
  • Faster incident response with real context

2) Headless agents in CI and containers

AgentSH

Problem

Unsupervised agents execute fast with no UI prompts. They fetch dependencies, run commands, and make network calls. A single unsafe step can become automated and repeated.

What to enforce

  • Deny-by-default network egress; allow only approved domains
  • Policy-based control of commands and subprocesses
  • Guard access to secrets and sensitive file paths
  • Redirect dependencies to internal mirrors and approved registries

Outcomes

  • Bound blast radius even when agents are wrong or tricked
  • Fewer "agent ran wild" incidents during automation
  • Deterministic policy enforcement with audit trails

3) Ops and deploy agents

AgentSH Watchtower

Problem

Teams are experimenting with agents that run terraform, kubectl, database actions, and incident tasks. These agents are powerful by default and operate in high-risk contexts.

What to enforce

  • Prompt or require approval for high-impact commands (deploy, delete, rotate, grant)
  • Restrict which clusters/accounts/environments an agent can touch
  • Lock down outbound network, especially to unknown endpoints
  • Central policies and emergency rollback controls

Outcomes

  • Safe automation without granting blanket admin privileges
  • Central control and approvals for high-risk operations
  • Clear accountability and auditability

4) Sensitive repositories and internal data

All products

Problem

AI tools and agents are increasingly pointed at codebases, tickets, docs, and data stores. Sensitive paths, tokens, and secrets are frequently within reach.

What to enforce

  • Prompt before reading secret-like files and sensitive directories
  • Prevent uploads to unknown destinations
  • Restrict tool access per repo/workspace sensitivity
  • Maintain centralized policy for different data classes

Outcomes

  • Least privilege for AI access by context
  • Reduced accidental leakage of credentials and IP
  • Evidence for compliance and security audits

5) Incident response: stop the bleeding

Watchtower

Problem

When something goes wrong, you need a fast, centralized response. Without a control plane, teams scramble to change configs or uninstall tools.

What to enforce

  • Fleet-wide kill switch for AI execution
  • Immediate policy tightening (deny-by-default mode)
  • SIEM forwarding and retention for investigation
  • Targeted allowlists to restore operations safely

Outcomes

  • Faster containment
  • Cleaner forensics
  • Controlled recovery
Getting started

Which product should I start with?

Beacon

If your AI runs on laptops/desktops, start with Beacon.

Learn about Beacon →

AgentSH

If your AI runs headless in CI/containers, start with AgentSH.

Learn about AgentSH →

Watchtower

If you need centralized policy, approvals, SIEM, and kill switch, add Watchtower (and scale across both environments).

Learn about Watchtower →

Tell us where your agents execute.

We'll help you prioritize the right starting point.

Prefer email? hello@canyonroad.ai
No spam. One email when it's real.