Standalone Mode

No LMS. No account. No server.

Write a specification, constrain the agent's tools, and run it against the test suite — every token generated on this machine's own GPU.

WebGPU · no backendWeights cached on-device0 B egress

Local runtime

Student specification

Tools granted to the agent

Load the model to enable the agent.

Agent trace

Plan and tool calls stream here during the run