Plans
Runtime security for AI agents — from solo builders to enterprise teams.
Basic
Solo builders running local agents
Monitoring
- Up to 500 events/day
- Process + filesystem monitoring
Controls
- Default block policies
- Custom policy rules
Logging
- 7-day log retention
- Log export
Support
- Community support
Pro
Power users with multiple tools and plugins
Monitoring
- Unlimited events
- Process, filesystem, and network
Controls
- Default + custom policy rules
- Approve/block/allow per tool
Logging
- 90-day log retention
- JSON + CSV export
Support
- Email support (48h SLA)
Team
Small orgs managing agent fleets + compliance
Monitoring
- Unlimited events
- Full stack + multi-agent
Controls
- Custom + org-wide policies
- Role-based access controls
Logging
- 1-year retention
- SIEM integration + API
Support
- Dedicated onboarding + SLA
Trust and data handling
Local-first
Scans run on your machine. Agent behavior is analyzed locally before any data leaves.
What is stored
Event metadata and risk scores for audit logging. Full event details when retention is enabled.
What is never uploaded
File contents, credentials, API keys, or any data flagged as sensitive by policy.
Export your logs
Download your full audit history as JSON or CSV at any time.
Frequently asked questions
What does Runtime Guard monitor?
Runtime Guard monitors three categories of agent behavior: shell commands and process spawning, filesystem reads and writes, and outbound network requests. Each action is scored against your configured policy.
Does it upload my data?
No. Scans and policy evaluation run locally on your device. File contents, credentials, and sensitive data never leave your machine. Only event metadata (timestamps, risk scores, action types) is stored if you enable cloud audit logging.
How does blocking work?
When an agent attempts a high-risk action (like reading SSH keys or making an outbound request to an untrusted domain), Runtime Guard intercepts the call and applies your policy: block, require approval, or allow and log.
What about false positives?
Policies are configurable. If a legitimate action is blocked, you can adjust your rules to allow it. The "Balanced" mode is designed for minimal false positives; "Strict" mode blocks more aggressively and is suited for high-security workflows.
Which runtimes are supported?
Currently, Runtime Guard works with any tool-using agent on your local machine. Direct integrations for Ollama, OpenClaw, and a "Tool wrapper" SDK are planned for v0.3–0.4. See the roadmap for details.
Can I try it before paying?
Yes. Run a demo scan to see the event timeline and risk scoring in action. Join the waitlist for early access to the full product.
Is there a free tier?
We're evaluating a free tier with limited event monitoring. During beta, waitlist members will get extended trial access. Join the waitlist to be notified.