Parser-free code index for AI agents cuts grep bills 42% with 4x smaller tool outputs

microbass · reddit · 2026-10-17

A developer released Reliary, an open-source offline Rust binary that indexes a repo into SQLite and serves it to AI agents over MCP — solving the context burn from grep-then-read cycles where every "where is X defined, who calls it" costs 3-4 billed round-trips.

The contrarian design: no parser

No tree-sitter, no AST, no per-language code; definitions and call sites come from structural detection on brace/indent/blank-line boundaries
One codebase, every language, same code path

Seven tools for the agent: findreferences (defs, callers, methods, dead code through one entry), search, callgraph, listmethods, finddeadcode, describe, plus a verify tool that checks a symbol-at-file:line claim against the index. A reliary wrap command prefixes verbose shell commands (cargo test, git diff) to compress output before it hits context.

Numbers: on a 10-question comprehension bench (4 seeds, claim-verified against the index, not keyword matching), it bills 42% less than grep with 4x smaller tool outputs and zero dead-ends, at comparable accuracy. The author explicitly does not claim higher accuracy than grep — the gap is within noise, with full methodology published plus an edit-outcome experiment that tied. The reproducible win is cost and output size.

MIT-licensed: cargo install reliary-agent or grab a prebuilt binary (linux/darwin/windows, x64+arm64); reliary trust . to index, reliary init wires up Claude Code / OpenCode / Pi / Cline.

Original post →

Parser-free code index for AI agents cuts grep bills 42% with 4x smaller tool outputs

More from coding & agent