Variable: MAX_SEARCH_QUERY_CHARS

const MAX_SEARCH_QUERY_CHARS: 512 = 512

Defined in: Services/Search/SearchQueryLimits.ts:26

Shared limit for user-supplied search queries.

This is a resource guard, not a usability limit — 512 characters is far more than any real search query.

Every semantic search path feeds the query straight into a transformer whose attention cost is O(n²). Passing truncation: true to transformers.js does not help: the feature-extraction pipeline already tokenizes with truncation on, and the bound it applies is the tokenizer's own model_max_length — 8192 for nomic-embed-text-v1.5. That leaves query length as a direct CPU and memory amplifier. Measured on a dev box:

| 50 chars | 19 ms | +0 MB | | 5,000 chars | 1.5 s | +218 MB | | 15,000 chars| 10.5 s | +1544 MB | | 60,000 chars| 58 s | +6680 MB |

15,000 characters is a perfectly legal GET under Node's default 16 KB header cap, so the ceiling has to be enforced in code.

Applied at three layers, deliberately: HTTP routes reject over-long queries outright (400), and each embedder clamps independently as a backstop for non-route callers.