GoKawiilPensieve local index cuts Claude Code tool calls and lookup cost by up to 43%
A new tool called Pensieve was benchmarked against plain grep-based search inside Claude Code across four real repositories ranging from 500 to 35,000 files. In held-out code-finding tasks, Pensieve located the correct file in 131 of 144 runs versus 123 of 144 for grep, while using 43% fewer tool calls, costing 18% less per lookup, and finishing 21% faster, with each search taking 0.1 to 0.2 seconds.
GoKawiil's interpretation of the reporting above, not reported fact.
The results suggest that local indexing could meaningfully cut the token and compute overhead of agentic coding tools like Claude Code, which rely heavily on repeated search and read calls to locate relevant files. Because savings reportedly grow with repository size, larger codebases could see the biggest efficiency gains if these measurements hold up in broader use. The benchmark appears to be self-reported, so independent verification would help confirm how the gains generalize beyond the four tested repos.
- Pensieve found the correct file in 131/144 runs vs 123/144 for grep alone across four repos of 500–35,000 files.
- It cut tool calls per lookup by 43% (2.0 vs 3.4) and cost per lookup by 18% ($0.041 vs $0.050).
- Lookups were 21% faster (8.6s vs 10.9s), with each Pensieve search taking just 0.1–0.2 seconds.
Source: tedhaley.ca, 2026-10-12
Published there as: “Local Indexing to reduce tool calling and token usage”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.