XERJ is a single‑binary, dependency‑free search engine that automatically indexes any directory tree using tree‑sitter to preserve symbols and line numbers, then exposes the data through a full Elasticsearch‑compatible REST API. In a controlled case study covering eight tasks across four languages, a coding agent that queried XERJ produced 9,982 output tokens ($1.58) and solved all 16 prompts, whereas a grep‑driven agent needed 26,477 tokens ($3.27) and a memory‑only agent required 260,916 tokens ($11.18) for only 11/16 solves – a 2.7× token reduction versus grep and a 26× reduction versus raw memory at identical solve rates. Field‑report usage shows roughly a 5× end‑to‑end token saving in everyday development. Indexing performance was measured on a sample corpus: 25,329 files processed in 158.1 seconds with eight workers, yielding 593 datasets, 83,103 live records and 790 junk entries. XERJ supports lexical, vector, hybrid and semantic search, agent‑level long‑term memory via /_memory namespaces, MCP‑based tool exposure, and optional reranking with an external Jev service. Benchmarks against Elasticsearch 8.13.4 show 1.72× higher ingest throughput, a 1.61× smaller on‑disk footprint, and 55 wins, 26 ties and 4 losses in a concurrent write‑read workload. Installation is a one‑line curl command; the --insecure flag disables TLS and API‑key authentication, intended for single‑user dev machines where the node binds only to 127.0.0.1 unless explicitly overridden.