Open-Source LearnHarness Turns AI Agent Loops Into a Smart Tutor

LearnHarness is an open-source desktop app that applies coding-agent architecture to learning difficult subjects. Its agents assess a learner’s current knowledge, map the prerequisites for a stated goal, teach through that map, and revise it as answers expose gaps. The GitHub repository has passed 5,700 stars at the time of writing, and the project ships under the AGPL-3.0 license .

An agent harness is the orchestration layer around an AI model. It controls prompts, tools, state, and the sequence of tasks the model may perform. LearnHarness applies that structure to tutoring, giving developers a concrete implementation of persistent learner state, bounded teaching stages, editable prompts, and bring-your-own-key model access.

Four stages, bounded jobs

Each session follows a fixed pipeline that limits what the agent does at each stage:

1. Measure. The agent probes the learner’s grasp of each prerequisite. Answers, confidence ratings, and written reasoning establish a starting point.
2. Plan. It builds a dependency map from the current knowledge state to the learning goal. The visible plan changes when later responses reveal a mistaken assumption or missing concept.
3. Teach. Each step starts from knowledge already recorded in the graph. The learner predicts an outcome, receives an explanation, and completes a comprehension check before advancing.
4. Practice. Separate sessions present varied problems that require retrieval without naming the relevant concept first. Problems and hints adapt to the learner’s recorded reasoning.