Limits

Declared Worker/API ceilings include: AI Search item and toMarkdown file ≤ 4 MiB; toMarkdown batch ≤ 16 files / 32 MiB input; per-file Markdown output ≤ 16 MiB; multi-instance request ≤ 10 instances; custom metadata fields ≤ 5; search results typically 1–50; context expansion 0–3.

Local quotas also bound items/chunks/vectors per instance, indexing jobs, provider concurrency, request/response bytes, and stream counts. Parser child defaults include a 30s deadline and bounded address/CPU/stderr. Live numbers:

ocd capabilities --json

These are not Cloudflare AI Search plan quotas. Full Workers AI inference limits do not apply because that surface is not provided.