Graft, an open-source project, has developed a new approach for AI agents that builds a project graph and only sends relevant nodes to large language models. This method significantly reduces token usage and processing time, showing a 42% decrease in tokens and a 60% time saving on the SWE-bench benchmark, while also improving correctness by 12 points. The tool aims to make AI agents more efficient by avoiding the need to process entire codebases for every prompt. AI
IMPACT This approach could significantly reduce the cost and increase the efficiency of AI agents that interact with large codebases.
RANK_REASON The cluster describes a new tool/framework for AI agents.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →