Economic Guardrails

Embed economic guardrails directly into AI systems to avoid uncontrolled executions

Start with our collection of guardrails

Browse a curated library of adaptive economic guardrails designed for common AI cost patterns: budget caps per execution, spend limits per tenant, token output constraints, model routing rules, cache-first policies, and more.

Each guardrail ispre-built and ready to deploy. Check our documentation to learn more.

Guardrails Library

Select guardrails to add to your rules

💰

Per-Tenant Cost Cap

Enforces a maximum spend threshold per tenant within a billing cycle. Pauses or throttles requests when the limit is reached.

Updated 2 weeks ago

COSTPER TENANTLIMIT

📉

Margin Protection

Monitors real-time unit economics and blocks operations when the cost-to-revenue ratio for a tenant exceeds a configured threshold.

Updated 1 week ago

REVENUEMARGINRATIO

🔁

Runaway Loop Detection

Detects agentic loops where an AI agent repeatedly calls tools or APIs without converging, cutting off execution before costs spiral.

Updated 3 days ago

AGENTICSAFETYDETECT

⚡

Token Budget Per Request

Limits the total LLM tokens consumed by a single agentic session or request chain, preventing unexpectedly expensive completions.

Updated 1 week ago

TOKENSPER REQUESTLIMIT

1from unomiq import UnomiqClient

3client = UnomiqClient(api_key="your-api-key")

5def run_agent_step(tenant_id: str, task: str):

6 # Check guardrail before each agent step

7 guardrail = client.guardrails.check(

8 tenant_id=tenant_id,

9 guardrail="per-tenant-cost-cap"

10 )

12 if guardrail.status == "exceeded":

13 raise RuntimeError(

14 f"Cost cap exceeded for {tenant_id}:

15 f"${guardrail.current_spend:.2f} / ${guardrail.limit:.2f}"

16 )

18 if guardrail.status == "warning":

19 print(

20 f"Warning: {tenant_id} at

21 f"{guardrail.usage_percent}% of cost cap"

22 )

24 # Proceed with the agent task

25 result = call_llm(task)

26 return result

Configure economic guardrails that fit your AI system

Every AI system has different economics: a customer-facing agent needs tight per-request limits, a batch pipeline needs daily budget caps, a multi-model workflow needs routing rules that balance cost against quality.

Configure guardrails at any level: per execution, per service, per tenant, or globally. Set hard limits that block requests when exceeded, or soft limits that log warnings and let the system degrade gracefully. Combine multiple guardrails into policies that reflect how your team actually thinks about cost.

Analyse the impact of using economic guardrails

See exactly how each guardrail is performing: how often it triggers, how much spend it prevented, and whether it's affecting the value your system delivers.

Compare periods before and after activation to measure real impact and identify guardrails that are too aggressive. Tune continuously based on data, not intuition.

Per-Tenant Cost Cap

Active

Invocations

Last 30 days

Triggered

2.7% trigger rate

Spend Prevented

Avg $24.65 per trigger

Daily triggers (30d)

avg 11.4/day

Recent triggers

customer_482exceeded

$127.402m ago

app_91warning

$42.185m ago

customer_203exceeded

$89.558m ago

user_1847warning

$31.0212m ago

customer_67exceeded

$204.3015m ago

FREE for developers, forever.

Signup and connect your telemetry and billing pipelines to start tracking unit economics across your AI systems in minutes.