how it works

One integration. Complete visibility. Total control over what your agents do.

Step 1: Setup

// one line. that's it.

import { AgentGuard } from 'agentguard'

const agent = AgentGuard.wrap(yourAgent)
logging enabled
cost tracking on
anomaly detection learning
budgets enforced
ready in 5 minutes. works with any agent framework.

Step 2: Protection in Action

see how agentguard protects you in real scenarios.

your agent is about to make a $200 API call.

1

agent requests action

"run gpt-4 batch analysis on 10,000 records"

2

agentguard evaluates

✓ cost: $200 (within budget)
✓ behavior: normal (matches baseline)
⚠ exceeds $50 approval threshold

verdict: requires human approval

3

you get notified

Slack • #agent-alerts • just now
🚨 Approval needed: data-processor-01 wants to run a $200 GPT-4 batch analysis on 10,000 records.
4

approved. executed. logged.

action completed successfully. full audit trail recorded:

✓ who approved: [email protected]
✓ when: Feb 4, 2026 at 2:34pm
✓ actual cost: $187.50
✓ result: success (10,000 records processed)

your agent just entered a retry loop. you're asleep.

agentguard learns your agent's normal behavior. when something's off, we catch it.

monday $5
tuesday $7
wednesday $4
thursday 3am $847
1

agent starts running

"process incoming data from webhook"

normal task. runs every hour.

2

something goes wrong

API returns malformed response. agent retries. and retries. and retries.

⚠ 2,341 API calls in 3 minutes
⚠ normal: ~10 calls per hour
⚠ cost climbing: $847 and rising
3

agentguard detects anomaly

ANOMALY DETECTED: retry loop

agent: data-processor-01

behavior: 2,341 API calls in 3 minutes

normal baseline: ~10 calls per hour

action taken: auto-blocked (kill switch)

4

crisis averted. you wake up to a report.

✓ agent stopped at $23.50 (not $1,100+)
✓ full log of what went wrong
✓ root cause: malformed API response at 3:02am
✓ money saved: $823.50

every action logged.
every anomaly caught.
every dollar tracked.

you're covered.

Secure Your Agent