Benchmark — vfs Saves 98.6% Tokens vs Reading Files

Self-benchmark results

Self-benchmark on this repository (pattern "Extract", 4,178 lines of source):

	Read all files	grep	vfs
Output size	101.9 KB	13.8 KB	1.5 KB
Lines	4,178	148	15
Est. tokens	26,079	3,537	373

• vfs saves 98.6% tokens vs reading all files (26,079 -> 373)
• vfs saves 89.5% tokens vs grep (3,537 -> 373)

Run your own benchmark

vfs bench --self                                   # self-test on vfs source
vfs bench -f HandleLogin /path/to/go-project       # benchmark on any project
vfs bench -f Login /path/to/project --show-output  # show actual output

The benchmark command compares three approaches on your codebase and prints a side-by-side table of output size, line count, and estimated token count.

Why the difference?

Reading all files returns everything — imports, comments, function bodies, blank lines. An AI agent processing this pays for every token, even though most of it is irrelevant.

Grep narrows it down to matching lines, but still includes partial function bodies, duplicate matches, and surrounding context that isn’t useful for discovering function locations.

vfs parses source via AST and returns only the exported signature — one line per function with the exact file and line number. No bodies, no noise.

Real-world impact

For an AI coding agent making 10 code searches per session, the difference compounds:

Method	Tokens per search	10 searches	Cost impact
Read files	~26,000	~260,000	High
grep	~3,500	~35,000	Medium
vfs	~370	~3,700	Low

Over a typical development session, vfs can save hundreds of thousands of tokens — which translates directly to faster responses, lower costs, and longer context windows for the AI.