How to use Claude to build a local idea finder
The author used Claude coding tools and other AI assistance to launch ahai, a local Mac app extracting ideas from Markdown files, while reporting active verification and known extraction errors.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Help turn [local Markdown extraction script] into a Mac app using [Python UI framework] and mlx_lm. Support source viewing, idea ordering, hiding, and exports. Identify high-risk changes for review and document extraction errors and processing limitations.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source poster / publisher
- rcanand2025
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reported launching ahai in ten days from concept, with local Markdown idea extraction and exports, while acknowledging rough edges and extraction errors.
- Human involvement
- The author explicitly described verifying everything, questioning suspicious output, and applying more diligence to high-risk parts.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- The author wrote a local MLX script to find ideas in a large Markdown collection.
- The author developed the script into a Mac application using Claude coding tools and other AI assistance.
- The author benchmarked local models and verified output while investigating suspicious results.
The author built ahai to recover ideas scattered across more than 13,000 Markdown files, including repository READMEs, Obsidian notes, clipped articles, and research abstracts. They first wrote a script using local MLX models on a Mac. Its results encouraged them to turn the approach into a product.
The application scans a selected folder for Markdown files, uses an mlx_lm model to identify and extract ideas, and gives each idea a title and short description. Users can open the rendered source, reorder or hide ideas, and export results as Markdown, HTML, or JSON. Processing supports pausing and resuming, although only one folder runs at a time.
The author named Claude Code, Claude Code on the web, and Claude Opus 4.5 for coding, alongside other AI tools for content and research. They described verifying everything, investigating suspicious output, and exercising greater diligence in high-risk portions. They also benchmarked local models to choose defaults appropriate to different machines.
The reported concept-to-launch period was ten days. The author acknowledged rough edges, false positives, false negatives, and processing delays.
Who it's for: This account may interest independent developers building local applications around searchable collections of personal Markdown material.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · rcanand2025
Frequently asked questions
Does the application send files to a hosted model?
The author describes ahai as a fully local Mac application using an mlx_lm model to process Markdown files.
What limitations did the author report?
Idea extraction has false positives and false negatives, folder processing takes time, and only one folder can run at a time. Processing can be paused and resumed.
More details
Remaining steps
- The author launched the application and documented extraction errors and processing limitations.
About this account
Claude tools assisted application development alongside other AI tools; the author retained verification, model evaluation, and product decisions.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=46120996. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: rcanand2025 on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Claude Opus 4.5 · Python · PySide 6 · mlx_lm
Local AI · Markdown · Mac app
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.