Image: GitHub
Why it mattersA developer picking up a new library can skip the generic explainer. The app finds where their understanding ends and starts the lesson there.
A chat window can explain a hard topic a hundred ways and still miss the one gap you need closed. LearnHarness tries a different shape. The open-source desktop app finds where your understanding ends. Then it plans a path to the goal and teaches you one step at a time. The repository opened on 16 October and has 1,115 stars nine days later. Trendshift ranked it 17 on its daily list on 24 October.
How the app works
The author describes four steps. First the agent measures. It asks questions and reads your answers and your written thinking. From that, it finds where your knowledge ends on each part the goal depends on. Then it plans. The agent draws a dependency map from your current knowledge to your goal. The plan is visible to you.
Then it teaches. Each step builds on something you already understand. You predict first, then read the explanation, then check your answer. If you misunderstood, that confusion is repaired before anything is built on it.
Last comes practice. Separate sessions pose varied problems. The problem must call for the right idea without naming it. Problems and hints change with your reasoning.
The knowledge graph carries across sessions
What a lesson teaches is stored as claims and the reasoning chains that lead to them. The author calls this a knowledge graph. Each node links back to the session where it was built. Future lessons can read the graph instead of starting from zero. Practice sessions can draw their material from it too.
You bring your own keys
LearnHarness has no subscription. You add an API key for your chosen provider. The author lists support for more than 30 provider options, including OpenAI, Anthropic, Google, DeepSeek and OpenRouter. A live cost meter shows what each call costs you.
The app runs on macOS, Windows and Linux. Chats, graph, settings and keys live on your machine. The author notes that model calls and web research still contact outside services. Local storage is not the same as offline. The system prompts are visible and editable.
For a team building software
A developer learning a new framework or a cloud API spends most of the time between two states. The reader already knows one part. A generic guide covers both. LearnHarness is a bet that the gap matters more than the explanation.
The project is one person's work so far: 193 forks, 70 open issues, AGPL-3.0 licence. The author sells no managed version. The GitHub repository holds a signed macOS build, a Windows installer, and a Linux AppImage. A reader can try the method today on a topic they are stuck on. The question for them is whether a measured starting point changes what they learn.
Source
Primary source: amosblomqvist/learnharness on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.