Skip to main content

​
Alert Types

Fishnet monitors for 10 different security events:

​
Prompt Drift

Type: prompt_drift
Severity: Warning/Critical Detects when prompts deviate significantly from established baselines using cosine similarity.
fishnet.toml
[llm.prompt_drift]
enabled = true
threshold = 0.85
baseline_window = 100

​
Prompt Size

Type: prompt_size
Severity: Warning Triggers when prompt length exceeds configured limits. 
[llm]
max_prompt_tokens = 8000

​
Budget Warning

Type: budget_warning
Severity: Warning Alerts when spending reaches 80% of daily budget.

​
Budget Exceeded

Type: budget_exceeded
Severity: Critical Triggered when daily spending limit is reached. New requests are blocked until the next day. 
[llm]
daily_budget_usd = 10.0

​
On-chain Denied

Type: onchain_denied
Severity: Critical Fires when a blockchain transaction is blocked by policy rules:
  • Destination address not whitelisted
  • Gas price exceeds maximum
  • Daily transaction limit reached
  • Value transfer too high

​
Rate Limit Hit

Type: rate_limit_hit
Severity: Warning Indicates the per-minute request limit was reached. 
[llm]
rate_limit_per_minute = 100

​
Anomalous Volume

Type: anomalous_volume
Severity: Warning Detects unusual spikes in request volume compared to historical patterns.

​
New Endpoint

Type: new_endpoint
Severity: Warning Alerts when an agent accesses a previously unseen API endpoint.

​
Time Anomaly

Type: time_anomaly
Severity: Warning Triggers for requests outside normal operating hours (configurable).

​
High Severity Denied Action

Type: high_severity_denied_action
Severity: Critical Fires when a high-risk action is blocked by policy (e.g., database deletion, credential exposure).

​
Alert Severity Levels

Alerts are classified into two severity levels:

Critical

Immediate threats requiring urgent attention:
  • Budget exceeded
  • On-chain transaction denials
  • High-severity denied actions

Warning

Potential issues to monitor:
  • Prompt drift
  • Budget warnings
  • Rate limits
  • Anomalous patterns

​
Configuring Alerts

Enable or disable alert types in fishnet.toml:
fishnet.toml
[alerts]
prompt_drift = true
prompt_size = true
budget_warning = true
budget_exceeded = true
onchain_denied = true
rate_limit_hit = true
anomalous_volume = true
new_endpoint = false
time_anomaly = false
high_severity_denied_action = true
retention_days = 30
Alert configuration can also be updated through the dashboard at /settings.

​
Alert Retention

Alerts are automatically cleaned up based on the retention policy:
  • Default retention: 30 days
  • Cleanup interval: Every 7 days
  • Startup cleanup: Runs on server start
To adjust retention: 
[alerts]
retention_days = 90

​
Webhook Notifications

Fishnet can send alerts to external services via webhooks.

​
Supported Providers

1

Create a webhook URL

  1. Go to Server Settings → Integrations → Webhooks
  2. Click “New Webhook”
  3. Choose a channel and copy the webhook URL
2

Configure in Fishnet

Via CLI:
# Store the webhook URL in the vault
fishnet add-key --service alerts.webhooks --name discord_url --key "https://discord.com/api/webhooks/..."
Via Dashboard:
  • Navigate to /alerts
  • Scroll to “Webhook Configuration”
  • Paste your Discord webhook URL
  • Click “Save”
3

Test the webhook

curl -X POST http://localhost:8473/api/webhooks/test \
  -H "Content-Type: application/json" \
  -d '{"provider": "discord", "message": "Test alert"}'

​
Webhook Payload Format

Fishnet sends alert notifications with the following structure: 
{
  "content": "Fishnet alert [critical] type=budget_exceeded service=openai time=2026-03-03T10:30:00Z message=Daily budget of $10.00 exceeded"
}

​
Webhook Retry Logic

Fishnet implements exponential backoff for failed webhook deliveries:
  • Max attempts: 3
  • Timeout: 8 seconds per request
  • Backoff: 250ms → 500ms → 1000ms
  • Retry conditions: 429 (rate limit), 5xx (server errors)
Webhook URLs must use HTTPS in production. Set FISHNET_DEV=1 to allow HTTP URLs during testing.

​
Webhook Security

Fishnet validates webhook URLs to prevent SSRF attacks:
  • ✅ Public HTTPS endpoints only (in production)
  • ❌ Localhost/loopback addresses blocked
  • ❌ Private IP ranges (10.x, 192.168.x, 172.16-31.x) blocked
  • ❌ Link-local addresses (169.254.x.x) blocked
  • ❌ Cloud metadata endpoints blocked

​
Alert Dispatch Behavior

Not all alerts trigger webhooks. The following types are sent:
Alert TypeWebhook Sent
prompt_drift✅
prompt_size✅
budget_warning✅
budget_exceeded✅
onchain_denied✅ (rate-limited to 1/hour)
rate_limit_hit✅
anomalous_volume✅
new_endpoint✅
time_anomaly✅
high_severity_denied_action✅
onchain_denied alerts are rate-limited to prevent webhook spam from repeated transaction attempts.

​
API Endpoints

​
List Alerts

GET /api/alerts?type=budget_warning&dismissed=false&limit=20&skip=0
Response: 
{
  "alerts": [
    {
      "id": "alert_123",
      "type": "budget_warning",
      "severity": "warning",
      "service": "openai",
      "message": "80% of daily budget consumed",
      "timestamp": 1709467800,
      "dismissed": false
    }
  ]
}

​
Dismiss Alert

POST /api/alerts/dismiss
Content-Type: application/json

{
  "id": "alert_123"
}

​
Get Alert Configuration

GET /api/alerts/config
Response: 
{
  "toggles": {
    "prompt_drift": true,
    "budget_warning": true,
    "budget_exceeded": true,
    "onchain_denied": true,
    "rate_limit_hit": true,
    "anomalous_volume": true,
    "new_endpoint": false,
    "time_anomaly": false,
    "high_severity_denied_action": true
  },
  "retention_days": 30
}

​
Update Alert Configuration

POST /api/alerts/config
Content-Type: application/json

{
  "prompt_drift": false,
  "retention_days": 60
}

Build docs developers (and LLMs) love

Get started for freeTalk to us