Antigma Labs 推出了 Ante，一款自包含的编码代理工具，以约 15MB 的 Rust 二进制文件形式交付，零运行时依赖，支持完全离线本地推理 [1]。该工具在 Terminal-Bench 2.1 基准测试中使用开源 DeepSeek V4 Flash 模型达到 82.7% 的性能表现 [1]，相比 Claude Code，其内存使用量降低约 7 倍，平均 CPU 使用降低约 9 倍，磁盘 I/O 降低约 5 倍 [1]。
Ante 内置推理引擎，可直接运行本地 GGUF 模型，无需 API 密钥或互联网连接 [1]。同时该工具支持 12 个以上的 LLM 提供商，包括 Anthropic、OpenAI、Gemini 和 Grok 等，实现无供应商锁定 [1]。项目部分代码开源，其中 SDK 和协议库采用 Apache 2.0 许可证发布，核心二进制在 Alpha 阶段采用 Binary Preview Terms [1]。
Antigma Labs has introduced Ante, a coding agent delivered as a self-contained Rust binary of approximately 15MB that requires no external dependencies and operates entirely offline [1]. The tool achieves 82.7% performance on the Terminal-Bench 2.1 benchmark using the open-source DeepSeek V4 Flash model, completing 368 out of 445 trials at a cost of roughly $68 [1]. Compared to Claude Code, Ante demonstrates substantially lower resource consumption, with memory usage reduced by approximately 7 times, CPU usage lowered by around 9 times, and disk I/O decreased by approximately 5 times [1].
The agent includes a built-in inference engine that enables local operation using GGUF models without requiring API keys or internet connectivity [1]. Ante supports more than 12 LLM providers, including Anthropic, OpenAI, Gemini, and Grok, avoiding vendor lock-in [1]. The project is partially open source, with the SDK and protocol libraries released under the Apache 2.0 license, while the core binary remains in Alpha under Binary Preview Terms [1].