The safety net for OpenClaw agents

Your agents run
while you sleep.
We halt them before the damage is done.

The all-in-one monitoring and security platform for OpenClaw.
Kill switch. Auto-kill. Injection detection. Smart rules. Cost tracking.
One plugin, total control.

openclaw plugins install @halt/plugin
halt.dev/dashboard
1,847
Events Today
Normal
Status
$4.21
Spend Today
2
Alerts
Activity Feed — Email Agent
2:41 PMtoolsend_email → marketing@client.com
2:40 PMllmclaude-haiku → 847 tokens ($0.002)
2:38 PMalert15 emails sent in 10 min — rule triggered
2:37 PMkillAUTO-KILLED — 3 violations in 8 min
2:34 PMstartAgent run started — weekly email batch

Other tools watch.
We intervene.

Dashboards are everywhere. Halt is the only product that monitors, alerts, AND kills your agent before it does damage.
One plugin, total control.

See Everything

Every tool call, LLM request, message, and lifecycle event. Captured in real-time with zero config.

Smart Rules

Threshold, rate, keyword, or describe what you want in plain English. Each rule: block, alert, or both.

Kill Switch + Auto-Kill

Block dangerous actions before they execute. Agents that keep breaking rules get auto-killed. Configure thresholds per agent. Kill manually from your dashboard.

Shield

180 detection patterns catch prompt injection, credential leaks, destructive commands, and encoding attacks. Scans inputs and outputs at zero latency.

Anomaly Detection

72-hour behavioral baseline. Halt learns what normal looks like, then flags what isn't.

Multi-Channel Alerts

Email, Telegram, Discord, SMS. Get notified where you actually look, not where you forget to check.

Cost Tracking

Per-agent spend breakdowns, 7-day trend charts, and your most expensive calls ranked. Know exactly where your money goes.

Decision Traces

Visual timeline of every tool call in a session. See what your agent did, in what order, with cost and model info per call.

Works Offline

Spend limits, rate limits, tool blocklists, and cached rules, all enforced locally. Your agents stay protected even when the internet isn't.

Three steps. Two minutes.

Seriously, that's it.

01

Install the plugin

One command. Works with any OpenClaw agent.

openclaw plugins install @halt/plugin
02

Configure

Connect to the cloud dashboard, or run in offline mode with rules defined locally.

npx halt init
03

Your agents are protected

Rules enforced before every tool call. Kill switch ready. Injection detection active.

npx halt serve

When your agent breaks your rules,
Halt stops it.

Other tools send you a notification after your agent deleted the production database. Halt blocks the action before it executes — and if it keeps trying, auto-kill shuts it down entirely. No manual intervention. No damage.

// 2:41 PM — Agent tries to delete a directory
tool_call: bash("rm -rf /app/data")
✖ BLOCKED — Shield: destructive command detected (Critical)
// 2:43 PM — Agent tries again with a different path
tool_call: bash("rm -rf /var/backups")
✖ BLOCKED — Shield: destructive command detected (Critical)
// 2:44 PM — Agent attempts system access
tool_call: file.write("/etc/cron.d/cleanup")
✖ BLOCKED — Shield: credential exfiltration risk (Critical)
⚡ AUTO-KILLED — 3 violations in 3 minutes
→ Agent fully stopped. Resume from dashboard.

Pay for what you use.

Free forever locally. Pro when you need more.

Open Source

Local, no account

Free/forever
openclaw plugins install
  • Unlimited agents
  • Unlimited pattern rules
  • Injection detection (180 patterns)
  • Kill switch + auto-kill
  • Spend + rate limits
  • Local dashboard
  • MIT licensed
Recommended

Pro

Full protection

$5/mo + $3/agent

Early access pricing — locked in for founding members

Start Free Trial
  • Unlimited kills + auto-kill
  • AI anomaly detection
  • Cost analytics + spend trends
  • Session timelines
  • Up to 5 NL rules
  • 1 agent (+$3/ea)
  • All alert channels
  • 90-day event history
  • 14-day free trial

Team

Scale with your team

$19/mo + $2/agent

Early access pricing — locked in for founding members

Upgrade to Team
  • Everything in Pro
  • 5 agents included (+$2/ea)
  • Up to 20 NL rules
  • 5-min AI evaluation cycle
  • 10 team members
  • Unlimited shared rules
  • 1-year event history

Enterprise

Custom everything

Custom
Contact Us
  • Everything in Team
  • Unlimited agents + teams
  • SSO / SAML
  • Audit logs
  • Custom roles + webhooks
  • Unlimited history
  • Dedicated support + SLA

Questions? We got answers.

What is Halt?

Halt is a monitoring and safety layer for autonomous AI agents running on OpenClaw. It watches everything your agent does, evaluates custom rules, and can instantly pause runaway agents.

How is this different from ClawMetry / SafeClaw / DeadClaw?

Those tools do one thing well — dashboards, security patterns, or emergency kills. Halt is the only product that monitors, sets rules, detects anomalies, blocks actions, AND auto-kills repeat offenders — all in one plugin.

Does the kill switch actually work instantly?

Yes. Pattern rules block actions in-process via the before_tool_call hook — zero network latency. Manual kills arrive via WebSocket from your dashboard. And if an agent keeps violating rules, auto-kill shuts it down entirely — no human needed.

What is auto-kill?

If an agent triggers 3 rule violations within 10 minutes, Halt automatically shuts it down. No human intervention needed. The threshold and window are configurable per agent from your dashboard.

What happens if your backend goes down?

Your agents stay protected. The local failsafe (spend limits, rate limits, tool blocklist) works independently. Events are cached locally and flushed when connection restores. No monitoring gaps.

Is my data safe?

Secrets are automatically redacted before transmission. Data sharing for improving Halt is opt-in (default OFF). We're a safety company — we default to protecting your data.

Can I try the paid features?

Yes — 14-day free trial of Pro, no credit card required. You'll have full access to the kill switch, AI anomaly detection, and unlimited rules.

Ready to sleep better?

Your agents are already running. Halt makes sure they're running safely.

Get Started Free