DigitalOcean Action Gateway Jev Tool
这是 Paddy Srinivasan 公开的项目资料。本站按原始来源展示项目信息，用中文说明适用场景和阅读边界；项目名、源帖与代码保持原样，便于逐项核对。
这条案例记录了什么
MCP 与 Skills、工具选择、Agent 与模型路由
以可调用工具的形式返回结构化判断。
社区公开项目或作者演示
原始来源：Paddy Srinivasan via X。作者自述，本站未独立复现。
Paddy Srinivasan
记录日期：2026-09-24。日期与身份应以原始资料为准。
怎样核对这个项目
- 先打开原始来源，确认作者、日期与 Jev 在项目中的具体用途。
- 如果提供仓库，再检查代码、运行要求和许可证；仓库存在不代表本站已经运行成功。
- 对速度、成本、准确率和规模数字，查看原文的任务、环境和计算口径。
- 真实调用入口和参数。
- 密钥及权限边界。
- 工具失败时的行为。
- 工具列表和版本。
原始文字与技术细节
以下内容保留原语言，供核对事实。中文页的场景说明是阅读提示，不是逐句翻译或实测结论。
展开英文项目摘要与原帖
项目摘要
Paddy Srinivasan says DigitalOcean exposes Jev both as a Serverless Inference model and as a first-class Action Gateway tool. The tool path lets agents discover Jev through a governed MCP endpoint and receive typed classifications and calibrated probabilities without a separate provider integration or prose parsing step.
来源原文
Jev from @typesafeai is now available two ways on @digitalocean : 1. As a model in our Serverless Inference fleet 2. As a first-class tool inside Action Gateway The second path matters. Any agent can discover and invoke Jev through the same governed MCP endpoint it uses for 16,000+ tools, without integrating another provider, managing another API key or parsing prose into a decision. Use the model when you want raw inference. Use the tool when your agent needs typed classifications, calibrated probabilities and auditable execution. The next phase of AI infrastructure is not just serving more models. It is turning specialized models into composable tools that agents can reliably use.
原记录的限制
- The post does not include a public repository or a runnable demo link.
- Availability and interface details may change with DigitalOcean's platform release.