· ai-devtools
AgentEnv 0.9.1334 enables running evaluations on Vercel Sandbox
Developers can now test agent performance using Vercel Sandbox, thanks to an update to Scale AI's AgentEnv.
AgentEnv, the framework designed to evaluate agents against specific tasks and score their outcomes, now supports Vercel Sandbox.
This Vercel provider is included starting with AgentEnv version 0.9.1334 and subsequent releases. The system operates by executing the agent and its associated task environment within distinct Sandboxes.
To see this in action, you can utilize the AgentEnv example to conduct an evaluation on Vercel Sandbox. This allows inspection of both the task's final score and the agent's recorded sequence of actions.
Once the setup guide is finished, execute a specific command from the repository you have cloned. This command leverages your established Vercel CLI login credentials.
Remember to substitute YOUR_TEAM with your specific Vercel team slug and YOUR_PROJECT with the name of an existing project.
Further details regarding this functionality can be found within the Sandbox ecosystem documentation.
