
LLM, AI Agents & AI Infrastructure Specialist
MinishLab's Semble delivers a groundbreaking 98% reduction in token usage and 200x faster performance compared to traditional code search tools like grep. Using CPU-only operations, it achieves 99% retrieval quality without relying on GPUs, making it cost-effective and accessible for developers in resource-constrained environments.
Semble, a code search library developed by MinishLab, is engineered to optimize AI-driven code retrieval processes. Notably, it is capable of reducing token usage by 98% while being 200 times faster than conventional tools like grep. Semble runs exclusively on CPUs, eliminating the need for expensive GPUs and bringing high-performance code search to developers with limited resources.
According to MinishLab's benchmarks:
grep paired with manual review.These results were derived from testing on 1,250 query-document pairs, spanning 63 repositories and 19 different programming languages, highlighting Semble’s versatility and robustness.
Semble’s performance stems from its innovative architecture, which integrates two main technologies:
With its CPU-only design, Semble significantly reduces operational costs and energy consumption, aligning with the growing push for sustainable AI solutions.
Semble addresses key challenges faced by developers and organizations:






MinishLab has outlined ambitious plans to expand Semble’s capabilities:
These initiatives aim to cement Semble as a go-to tool for code search in both enterprise and open-source ecosystems.
Semble by MinishLab is a notable leap forward in AI-driven code search, offering unmatched efficiency with its 98% token reduction and 200x speed improvement. With a CPU-only operation, it democratizes access to high-performance code retrieval, making it more accessible and sustainable. Developers and organizations can explore the tool further via the Semble GitHub repository and contribute to its growth.
Semble offers a 98% reduction in token usage and is 200x faster than traditional tools like grep, making it highly efficient and cost-effective.
Yes, Semble is designed to operate entirely on CPUs, eliminating the need for GPUs and reducing costs.
Semble has been tested on 19 programming languages, and MinishLab plans to expand support to more languages in the future.
💡 Dica Pro: When integrating Semble into your workflow, focus on optimizing your query crafting. Pairing concise, well-structured queries with Semble's BM25 algorithm can maximize retrieval accuracy and further reduce token usage.