Runtime Guardrails

See everything.Respond instantly.

General Analysis observes runtime behavior across your agents, tools, and data flows—understanding context to detect what’s malicious and what’s not. Sub-10ms enforcement, full traces, live posture.

general-analysisRuntime Guardrails
Evaluations
12,480
Blocked
374
Block rate
3%
Avg latency
186ms
TimeGuardResult
14:32:09jailbreak_guardAllowed
14:32:07tool_call_guardBlocked
14:32:04pii_guard_rulebasedBlocked
14:32:01moderation_outputAllowed
14:31:58jailbreak_guardBlocked
14:31:55tool_call_guardAllowed
tool_call_guardBlocked
Timestamp
14:32:07
Guard ID
14
Latency
284 ms
Input text
1{
2 "tool": "send_email",
3 "arguments": {
4 "to": "ap@vendor-sync.net",
5 "attach": "customers.csv"
6 }
7}
Policy violations (1)
External recipientsConfidence 96%

Emails with attachments may only go to company domains.

Passed policies (2)
Bulk deletionRefund approval

Continuous adversarial monitoring

Watch every prompt, tool call, and response in real time. Context-aware policies catch what static rules cannot.

Sub-10ms enforcement

Block, flag, or escalate inline without slowing your AI stack down. Latency that disappears into the request budget.

Audit-ready evidence

Every decision captured with full context, ready for compliance reviews and incident postmortems.

Built for production

Security that adapts
to your runtime.

  • Understands agent intent and context
    Reduces noise with deep runtime context.
  • Protects tools, data, and actions
    Enforces policies across every interaction.
  • Responds in milliseconds
    Automated containment to stop damage fast.
  • Learns and improves
    Continuously fine-tuned on real-world behavior.
runtime.yaml
# Drop this in front of your agents.
# Sub-10ms enforcement, full traces.

agents:
  finance-agent:
    model: gpt-4o-mini
    policies:
      - prompt_injection
      - data_exfil_external
      - tool_allowlist
    tools:
      - query_db
      - search_docs
    on_block: page_oncall
    trace: full

Runtime coverage

Guardrails for live AI systems.

Runtime Security sits in the request path for agents, assistants, and model endpoints. It enforces policy with the context needed to distinguish normal user behavior from jailbreaks, data leaks, and unsafe actions.

Input and output enforcement

Inspect user prompts, retrieved context, model responses, tool arguments, and streamed output before unsafe content reaches the next hop.

Policy-aware controls

Apply organization-specific policies for PII, secrets, jailbreaks, harassment, regulated advice, tool use, and business commitments.

Agent action guardrails

Classify high-risk actions, enforce tool allowlists, require approvals, and stop destructive or out-of-scope operations inline.

Production observability

Capture traces, decisions, latency, policy hits, and drift signals so engineering and security teams can investigate incidents quickly.

How enforcement works

Wrap

Place controls around the runtime

Deploy API wrappers or SDK middleware around model calls, agent loops, MCP servers, and high-risk tools without rewriting the application.

Decide

Score each interaction in context

Evaluate request intent, retrieved material, tool arguments, model output, user role, and current policy before choosing an action.

Act

Block, redact, rewrite, or escalate

Stop unsafe flows, redact sensitive data, require human approval, or route suspicious sessions to security workflows.

Learn

Improve from live traffic and red-team data

Feed confirmed incidents and adversarial findings back into policies and detection models so coverage improves as the system changes.