Tag
Maka has built-in support for DeepSeek API's web search. By declaring the web_search tool in the API request parameters, it eliminates the need to integrate third-party search engines, further improving compatibility with models like DeepSeek/GLM/Kimi.
On the Kimi K3 model, the open-source agent framework Maka achieves a 10% higher overall pass rate on Terminal-Bench 2.1 compared to the official Kimi Code CLI, and 20% higher on hard tasks. Through optimizations like context-budget pruning, streamlined tool surface, and concise system prompts, significant performance gains are realized. The full report and harness are open-sourced.
The Maka project is undergoing a comprehensive pre-release review and has started compiling design documents, aiming to help developers learn Agent development. It is a local-first Agent workspace with 640 stars and an active community.
Maka's Harness project improved the self-check mechanism, enabling DeepSeek Flash V4 to achieve evaluation results close to GLM-5.2 on the terminal-bench sample set, completing 10 programming agent tasks with only 4 RMB and a 97.5% cache hit rate.