Policy

Runtime Policy

Guardrails that keep AI Dev Hub fast and reliable for builders: clear rate budgets, deterministic validation, and operator safety controls.

Runtime Rate Budgets

Risk Class Requests Window Input Cap
L (light transforms) 60 60s 128 KB
M (richer transforms) 30 60s 256 KB
H (network-sensitive) 10 60s 512 KB

Runtime limits are per identity and return HTTP 429 with retry headers when exceeded.

Payload + Execution Limits

  • Request body cap: 1 MB
  • Response body cap: 1.4 MB
  • Max concurrent requests per identity: 6
  • Timeout envelope: L=2000ms, M=4000ms, H=8000ms
  • Default daily network budget: 3000 network calls (operator configurable)

Dedicated API Limits

Dedicated API tools run on their own endpoints and currently follow the free-tier budget below.

  • Per-day budget: 50 requests/day per IP
  • Burst budget: 10 requests per 10s window

Safety Guardrails

  • Private, loopback, and local-network targets are blocked for runtime network fetches.
  • Input schemas are validated before execution to prevent malformed tool calls.
  • Operator kill switches can pause runtime or network budgets during incidents.
  • Structured runtime logs are emitted for rate-limit and failure monitoring.

Error Code Quick Reference

Code Meaning
400 Invalid input, wrong surface, or browser-only tool on runtime endpoint
429 Rate-limited or daily network budget exhausted
503 Runtime paused by operator policy
504 Upstream/network timeout