Guardrails

Safety controls and policy enforcement for all agents

🔒

PII Filter

Detect and mask personally identifiable information before sending to LLM

234 blocked / 12,500 total

💰

Cost Limit

Block requests that would exceed per-run cost threshold ($0.50 default)

12 blocked / 12,500 total

🛡

Content Safety

Block harmful, violent, or inappropriate content in inputs and outputs

89 blocked / 12,500 total

☣

Toxicity Filter

Score output toxicity and block responses above threshold (0.7)

📏

Output Length

Limit maximum output tokens to prevent runaway generation

45 blocked / 12,500 total

🔄

Loop Detection

Detect and break infinite tool-calling loops after N iterations

3 blocked / 12,500 total

⏱

Timeout Guard

Kill runs exceeding maximum duration (60s default)

18 blocked / 12,500 total

🏷

Topic Restriction

Restrict agent to approved topics only (configurable per agent)

📊

Hallucination Check

Cross-reference outputs against knowledge base for factual accuracy