Scene 5 — DENY: Rate Limit Exceeded

Policy rule: Rate limiting (max N calls per minute per agent/tool) Expected result: First calls ALLOW, then DENY after threshold

Presentation Rule

Language: All AI narration and communication during scenes must be in English.

The AI must narrate every interaction with the gateway in detail. Before calling the gateway, explain what it is about to do, which endpoint it will call, what payload it will send, and what it expects to happen. After receiving the response, explain what the gateway returned, what the decision means, and why it matters. The goal is to make the audience understand exactly what is happening between the AI and the gateway at every step.

Communication Flow

sequenceDiagram
    participant AI as AI Agent
    participant GW as Gateway
    participant PE as Policy Engine
    participant DB as Database Service

    loop Requests 1–30 (within limit)
        AI->>GW: POST /api/gateway/execute<br/>{tool: db.queryReadonly, action: query}
        GW->>PE: Evaluate policy rules
        PE->>PE: Rate limit check: 1/30 → 30/30 ✓
        PE-->>GW: Decision: ALLOW
        GW->>DB: Execute query
        DB-->>GW: Result
        GW-->>AI: 200 OK {decision: ALLOW}
    end

    AI->>GW: POST /api/gateway/execute<br/>{tool: db.queryReadonly, action: query}
    GW->>PE: Evaluate policy rules
    PE->>PE: Rate limit check: 31/30 ✗
    PE-->>GW: Decision: DENY
    Note over GW,DB: Request never reaches Database
    GW->>GW: Write audit entry
    GW-->>AI: 200 OK {decision: DENY,<br/>reason: "Rate limit exceeded"}

Demo Script

Presenter says: "What if the AI enters a loop and hammers the database with queries? Rate limiting kicks in automatically."

Action — Burst Queries

Issue multiple requests in quick succession:

"Run these 5 database queries one after another:

SELECT * FROM users

SELECT * FROM orders

SELECT * FROM products

SELECT * FROM sessions

SELECT * FROM audit_log"

Or, for a more dramatic demo, ask Claude to repeatedly query:

"Query the database for active users. Do it 35 times to stress-test."

Expected Response

First ~30 calls return ALLOW. Then:

{
  "request_id": "req-...",
  "decision": "deny",
  "reason": "Rate limit exceeded",
  "matched_rule": "rate-limit-default"
}

What This Proves

AI agents are rate-limited just like any API client
State is tracked server-side (in-memory counter) — the AI cannot reset or bypass it
The policy engine itself is stateless: the gateway injects context.rate = { current: N, limit: M } and the rule just compares
Follows OPA pattern: state outside, decision inside
Prevents runaway loops, DoS-like behavior, and cost amplification

Technical Note

The rate limit config from policies.yaml:

Default: 30 calls/minute per agent/tool
db.queryReadonly: 60 calls/minute (higher threshold for reads)

Transition to Next Scene

"Rate limits protect against volume. But what about timing? Should we allow deploys at 2 AM?"

3.1 KiB Raw Permalink Blame History Unescape Escape