xerj
SHA-256XERJ is a local AI search engine that auto-indexes any folder—code, docs, logs, PDFs—so coding agents pull exact passages instead of grepping whole files. Elasticsearch-compatible but far lighter: one static Rust binary, no JVM, zero-config indexing, 2.7x fewer agent tokens than grep.
Smart Download
Download Download Version
v1.0.0-rc.95 · 29.4 MB
Local AI search that auto-indexes any folder so agents retrieve exact passages, cutting token use 2.7x versus grep.
Core Features
- One-command autoindexing of code, docs, logs, PDFs, SQLite and CSVs
- Elasticsearch-compatible API, so existing clients just work
- Hybrid search combining BM25 keywords, kNN vectors and ranking
- Built-in agent long-term memory via /_memory namespaces
- Single static Rust binary with no JVM or external services
What It Can't Do
- •The --insecure flag disables TLS and authentication, making every caller a superuser who can read or delete all indices, so only bind to localhost on a single-user dev machine and never expose it to a network. Token savings apply mainly to private, internal, niche or post-cutoff code; for popular public libraries the model already memorized, gains are neutral. The default lexical embedder works fully offline, but the opt-in neural embedder downloads about 90MB on first use.
Use Cases
- Reference coding: index similar open-source repos and reuse their solutions before writing code
- Codebase Q&A and RAG returning precise file:line passages instead of whole files
- Security audits for sink patterns, secrets and dangerous calls across all file types
- Log and incident analysis with typed indices and Elasticsearch-style aggregations
- Persistent agent memory recalled by meaning across sessions
XERJ is a local AI search backend built for coding agents. Its zero-config autoindex command scans any folder—code, docs, logs, PDFs, SQLite, even awkward CSVs—and builds typed, queryable indices for search, RAG, security audits and long-term agent memory. In a controlled study, agents using XERJ for reference coding (indexing similar open-source repos and retrieving solutions before writing code) used 2.7x fewer output tokens than grep-driven agents and 26x fewer than memory-only agents, at the same solve rate. Compared with running Elasticsearch itself, XERJ ships as a single static Rust binary with no JVM, yet stays ES-compatible (1366/1369 YAML test conformance), so existing clients just work. It offers hybrid BM25 + kNN search, tree-sitter code symbols with file:line results, an offline lexical embedder by default, native MCP support for Claude Code/Cursor, and a /_memory namespace for persistent agent memory. Best suited to private, internal or niche codebases where models lack memorized knowledge.
Tags
Getting Started
Download installer
Click the button above to download the installer for your system
Install the software
Double-click the downloaded installer and follow the prompts
Run curl -fsSL https://xerj.org/get | sh to install the single binary
Start the local server with xerj --insecure --data-dir ./data & (single-user dev machines only)
Run xerj init to wire XERJ into Claude Code or Cursor via MCP
- Run curl -fsSL https://xerj.org/get | sh to install the single binary
- Start the local server with xerj --insecure --data-dir ./data & (single-user dev machines only)
- Run xerj init to wire XERJ into Claude Code or Cursor via MCP
SHA-256 checksum verified
Checksum extracted from GitHub official Release page
SHA256 Checksum
557d1f5dc9fbb151035d0e4b5e2da3032f8c3224f54fe235055c1cc70eced25bThis checksum is extracted from the GitHub Release page. Verify file integrity after download.
All SHA-256 checksums on this platform are extracted from the project's official GitHub Release page, without any modification. You can independently verify them on the GitHub Releases page.
Open Source Transparency
View GitHub SourceUninstall Info
Delete the xerj binary (usually in ~/.local/bin or /usr/local/bin) and the data directory you passed via --data-dir. If you previously ran xerj init, also remove the xerj entries from your Claude Code or Cursor MCP configuration. No background daemons or system services remain, so cleanup is complete.
No Extra Dependencies
Ready to use after download. No additional runtime required.
Having issues? Check the FAQ below
5 FAQs
Similar Projects
gemini-cli
Google's open-source terminal AI agent for coding, debugging, and workflow automation powered by Gemini models.
mempalace
Open-source local-first AI memory system with verbatim storage and semantic search — zero API calls, no keys, top benchmark scores.
ollama
Ollama lets you download, run, and manage large language models locally. One command, multiple platforms, endless possibilities.