Build a local knowledge app for agent outputs with Codex
Hey everyone,. Task: Hey everyone, I wanted to share something I built using Codex 5.5 Pro. I created it after spending the last 12 months working with agents like Codex and OpenClaw on my MacBook and Mac mini. I had files all over the place and needed a way to centralise the information my agents were producing, so I developed a local knowledge app, called Mianotes, that your agents can use to document the work they are doing. It's free and open source. Mianotes can also convert video and audio files, PDFs, images, and links, including YouTube links, into Markdown. It uses Microsoft MarkItDown for file conversion and OpenAI or compatible APIs to improve OCR when MarkItDown needs help. The platform, APIs, MCP server, dashboard, macOS and Ubuntu installers, documentation, and marketing site were all created in just 2 weeks, from the first commit to the first macOS package. I used 4 Codex agents and more than 30 agent sessions to build it. The Codex agents also documented the product inside Mianotes as they worked, which gave me a way to test the idea while building the product. Here's the first public version of the app: → Website: https://mianotes.com/ → Docs: https://mianotes.com/docs/ → GitHub: https://github.com/Mianotes If you try it, I'd love to know what works, what doesn't, and what you'd expect from a tool like this. Thanks; tools: Codex. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include access to Codex, OpenAI APIs, and Microsoft MarkItDown. Integration uncertainty exists regarding API costs and OCR accuracy for complex documents. Source reliability is low as it is a self-reported Reddit post with no independent verification. Human approvals are required for final code commits and public release. Proposed advice: Test API rate limits and OCR quality on diverse file types before scaling..
Tools
Published on Agent Work Atlas:
Overview
Hey everyone,
Workflow
Hey everyone, I wanted to share something I built using Codex 5.5 Pro. I created it after spending the last 12 months working with agents like Codex and OpenClaw on my MacBook and Mac mini. I had files all over the place and needed a way to centralise the information my agents were producing, so I developed a local knowledge app, called Mianotes, that your agents can use to document the work they are doing. It's free and open source. Mianotes can also convert video and audio files, PDFs, images, and links, including YouTube links, into Markdown. It uses Microsoft MarkItDown for file conversion and OpenAI or compatible APIs to improve OCR when MarkItDown needs help. The platform, APIs, MCP server, dashboard, macOS and Ubuntu installers, documentation, and marketing site were all created in just 2 weeks, from the first commit to the first macOS package. I used 4 Codex agents and more than 30 agent sessions to build it. The Codex agents also documented the product inside Mianotes as they worked, which gave me a way to test the idea while building the product. Here's the first public version of the app: → Website: https://mianotes.com/ → Docs: https://mianotes.com/docs/ → GitHub: https://github.com/Mianotes If you try it, I'd love to know what works, what doesn't, and what you'd expect from a tool like this. Thanks
Reported result
Actual outcomes are unknown as the source provides no measured results or performance metrics. The author claims the app was built in 2 weeks and is free/open-source, but no data on adoption, stability, or conversion accuracy is provided. Proposed acceptance check: Verify that a sample PDF and YouTube link are correctly converted to Markdown with >90% text fidelity and that the macOS installer launches without errors.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include access to Codex, OpenAI APIs, and Microsoft MarkItDown. Integration uncertainty exists regarding API costs and OCR accuracy for complex documents. Source reliability is low as it is a self-reported Reddit post with no independent verification. Human approvals are required for final code commits and public release. Proposed advice: Test API rate limits and OCR quality on diverse file types before scaling.
Try it
Act as a senior developer. I want to build a local knowledge app called Mianotes using Codex agents. Inputs: [replaceable: list of sample files including PDF, MP4, and YouTube URL], [replaceable: target OS (macOS/Ubuntu)], [replaceable: API keys for OpenAI]. Steps: 1. Initialize a project structure for a local-first app. 2. Implement file conversion logic using Microsoft MarkItDown for PDFs and images. 3. Integrate OpenAI API for OCR enhancement on low-quality scans. 4. Create a module to extract transcripts from video/audio and YouTube links. 5. Generate a macOS installer package. 6. Document the product within the app itself using agent-generated notes. Output: A working prototype codebase and a README.md. Quality Measure: Run a dry run on one sample PDF and one YouTube link. Verify Markdown output is readable and accurate. Safeguards: Do not publish or spend API credits without my explicit confirmation. Ask clarifying questions if API keys or specific file formats are missing.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.