Skip to main content
Limits lets you check (evaluate) requests (for example, what your AI agent is about to do or has returned) and get back allow, block, or escalate. You use the Limits SDK or API in your app: you call with a policy key and a payload; Limits evaluates and returns the outcome. When you define a policy in the platform, you choose one of two modes: Conditions or Instructions. They are two different ways to define how that evaluation works. You do not use both in the same policy—each policy is either in Conditions mode or Instructions mode.

​
How you use Limits: check (evaluate) with the SDK

In your application you call the Limits SDK to evaluate (check) a request. You pass the policy key and the request payload (e.g. amount, user, resource). Limits evaluates and returns allow, block, or escalate. 
const decision = await limits.check('your-policy-key', {
  amount: 100,
  user: { risk_level: 'low' },
});

if (decision.isAllowed) console.log('Allowed');
if (decision.isBlocked) console.log('Blocked');
if (decision.isEscalated) console.log('Escalated');
So the platform is for checking/evaluating the response or context from your AI agent (or any request you want to govern) and acting on the result: allow, block, or escalate.

Conditions

Deep dive into conditions (operators, field paths, SDK/API).

​
Quick difference between the two modes

Conditions modeInstructions mode
AnswersWhen or if something appliesWhat to do (required action or behavior)
RoleRules that decide allow / block / escalateTells the system what to do; used to evaluate and return allow / block / escalate
FormatStructured rules (e.g. field, operator, value)Natural language (text, bullets, steps)
  • Conditions = “Under what circumstances does this apply?” — you define rules (e.g. “block if amount > 10000”).
  • Instructions = “What should happen?” — you describe the required action or behavior in plain language.
You pick one mode per policy. Conditions and instructions are not used together in the same policy. Conditions vs Instructions mode in the policy editor

​
Conditions mode: When or if something applies

In Conditions mode you define when a request is allowed, blocked, or escalated using structured rules. You define things like:
  • Block if request.amount > 10000
  • Escalate when user.risk_level == 'high'
  • Allow when request.resource is in the allowed list
So in Conditions mode you answer: “When does this rule apply?” or “If this is true, what outcome?”

​
Example (Conditions mode)

Policy in Conditions mode: block high-value payments, escalate risky users.
  • Block when: request.amount > 5000
  • Escalate when: user.risk_level == 'high'
  • Allow when: neither of the above matches
When you call the SDK with { amount: 100, user: { risk_level: 'low' } }, Limits evaluates these rules and returns allow, block, or escalate.

Conditions

Build conditions with operators and field paths.

​
Instructions mode: What to do

In Instructions mode you describe what should happen—the required action or behavior—in natural language. The system uses that to evaluate requests and return allow, block, or escalate. You might write, for example:
  • “Block requests over $10,000.”
  • “Escalate when the user is high risk.”
  • “Allow only if the resource is in the approved list.”
So in Instructions mode you answer: “What is the required action or behavior?” in prose; the platform uses that to decide the outcome.

​
Example (Instructions mode)

Policy in Instructions mode: you write instructions such as:
  • “Block any payment over $5,000.”
  • “Escalate when risk level is high.”
  • “Otherwise allow and log the request.”
When you call the SDK with a payload like { amount: 100 }, Limits evaluates using these instructions and returns allow, block, or escalate.

Instructions

Full guide to instructions policies.

AI Assistant

Create policies from natural language with the Assistant.

​
Summary

  • How you use Limits: Call the SDK or API to check/evaluate a request (e.g. from your AI agent); you get back allow, block, or escalate.
  • Conditions mode = When or if something applies. Structured rules that decide allow / block / escalate. One way to define a policy.
  • Instructions mode = What to do. Natural language that describes the required action or behavior; the system uses it to evaluate and return allow / block / escalate. A different way to define a policy.
  • You choose one mode per policy. Conditions and instructions are not used together in the same policy.

​
Where to edit in the platform

  • Conditions mode — Policies → select a policy → Conditions (or /policies/[id]/edit/conditions). Define when to allow, block, or escalate with the condition builder.
  • Instructions mode — Policies → select a policy → Instructions (or /policies/[id]/edit/instructions). Write what should happen in natural language.
You can switch between the two modes from the policy edit page; each policy has one mode.

Conditions

Conditions deep dive.

Instructions

Instructions deep dive.

Guardrails

Guardrails (PII, moderation, etc.).

Platform: Policies

Create and manage policies in the UI.