Benchmarks

O(1) vs O(n) context efficiency.

Reproducible measurements comparing one registry tool context versus one wrapper per CLI.

Catalog Size

845

CLIs benchmarked

O(1) Tokens

Per-query context floor

Discover p95 (captured)

1574ms

Observed during benchmark run

Generated

19d ago

API source: https://api.clime.sh · refresh recommended

Methodology

Single registry tool context (`clime.search`) regardless of catalog size.
One wrapper tool per CLI provided directly to the agent context.
Estimated tokens = UTF-8 bytes / 4 (rounded up).
O(1) latency measured from live /v1/discover API calls. O(n) latency measured as local linear scan over in-context tool wrappers.

Comparison Table

Tool Wrappers	O(n) Tokens	O(1) Tokens	Ratio (O(n)/O(1))	Reduction with O(1)	O(n) Local scan ms
10	939	74	12.69x	92.12%	0.0132
25	2272	74	30.70x	96.74%	0.0211
50	4636	74	62.65x	98.40%	0.0375
100	9229	74	124.72x	99.20%	0.0708
200	18689	74	252.55x	99.60%	0.1345
500	46098	74	622.95x	99.84%	0.3776
845	77121	74	1042.18x	99.90%	0.6662

Reproduce locally with `npm run benchmark:context`.