Local web search tool says it cuts token use 87% and cost 66%
Remote-Breadfruit204 · reddit · 2026-07-23
A local web search stack claims 87% fewer tokens and 66% lower cost
The author built webfetch as an alternative to hosted search from Anthropic and OpenAI, arguing that the default model is expensive: $10 per 1k searches plus roughly 17k result tokens per query.
Key pieces of the system:
- RRF across 4 search engines, local page fetching, BM25 + bi-encoder retrieval, and a cross-encoder reranker
- Sentence-level compression that halves result tokens without measured recall loss
- Semantic caching with embedding match + NLI verification and adaptive TTLs
- Provenance shown for cached results, with the model able to force a fresh search
On its SimpleQA benchmark, the author says the same agent loop matched hosted search at 96% accuracy while using 87% fewer tokens and costing 66% less. One test loop of 16 searches reportedly saved $1.50. It ships as a PyPI install and can be added to Claude Code as an MCP server.
Related event: Open-source Webfetch slashes AI agent search costs and tokens(4 posts)→
More from coding & agent
- Claude Code finds root backdoor in Honda Civic head unit, then debloats it — IndraVahan · 2026-10-26
- Claude Code + Monid MCP builds a $12K/year B2B prospecting engine on live data — aryanXmahajan · 2026-10-26
- Karpathy shares CLAUDE.md rulebook to stop AI coding agents from piling up tech debt — mdancho84 · 2026-10-26
- Rethinking dependencies in the agent era: drop libs that save 20 lines of code — gethackteam · 2026-10-26
- RustGIS hits 2,000 GitHub stars in two weeks with agent-first architecture — giswqs · 2026-10-26
- Vercel Ships 12 New Products at Its 2026 Conference as It Races Rivals for the Agent Entry Point — FinanceYF5 · 2026-10-26