DOCS / OVERVIEW
A learning subagent. Its own harness.
Your coding host runs the AI. Learning Harness runs a separate learning loop through that existing connection, using selected context from the work.
The coding agent delivers your task. The learning agent helps you understand it. It has its own tools, local learner context, pacing and assessment, so teaching does not become another responsibility in the coding prompt.
No Learning Harness model service. No Learning Harness backend receiving your code. Learning requests use your host’s existing model connection and allowance. Memory stays local. Selected work still reaches that model provider; anonymous usage metadata is optional and separate.
Start with your coding host
Codex Plugin, hook review and browser panel →Claude.app Code Local session and subscription bridge →OpenCode Desktop 2.2.5 plugin and side panel →
Understand the boundaries
Architecture explains the two independent loops. Privacy traces the data. Token usage explains the additional learning cost.
Desktop beta installers are available from the product site. These guides describe the local desktop distribution. No public npm package or marketplace listing is available yet.