I’ve been wrestling with the app’s core problem: it tries to do too much on its own, but its real strength is in integrating with tools like Obsidian and Notion. That’s where the value lies, but I’m still figuring out how to communicate that clearly to users. A Reddit comment praised the local transcription and the no-subscription model, which felt good to see, but it got removed, so I’m not sure how much traction that kind of feedback will get.

I decided to skip the beta phase entirely, removing the beta feedback prompts and reworking the contributor thank-you section. The feedback forum is still open for bug reports and future ideas, but I’m relying mostly on existing feedback to polish things. I’m also shifting focus toward GameDev communities since they might get more out of the integration features. To help with that, I’m planning to make short video demos showing how the app fits into actual workflows.

On the technical side, I did a big refactor of AppState and Control Center using Codex, breaking the code into smaller files and cutting the lines from around 2-3k down to about 500. I branched this refactor commit to keep the main codebase safe while I test it out. The app launches and runs pipelines fine now, but it still doesn’t abort cleanly if you stop mid-transcription or mid-summarization, which is a frustrating bug I need to fix.

I upgraded the app to use OpenAI models 4.0+ instead of 3.0, and tested Mistral 7B and Qwen 3 models. Qwen 3 gives better quality but is slower, so I set up auto-optimization based on VRAM: if the system has more than 8GB VRAM, it uses Qwen 3; otherwise, it falls back to Mistral. I also updated the Ollama model list and tested Qwen 3 as a drop-in replacement, but it’s slower than the others. Documentation on model performance and pricing is still pretty sparse, which makes it hard to know if I’m making the best choices.

With the release deadline looming on May 15th—about eight days away now—I’m prioritizing polish and stability improvements. I’m also considering a Cursor subscription to help with code refactoring and repo analysis, though I’m not sure yet how effective that will be. On a related note, I’m exploring local codex-like tools for game dev or small teams, inspired by Watchlight Studio’s offline approach, but I don’t know how feasible that is or if anything suitable already exists.

The big open questions for me right now are how to fix the abort issue during transcription and summarization, how to better communicate the app’s integration-focused value, and whether skipping beta and targeting GameDev communities will actually help with adoption. I’m also unsure if the current model setup is the best balance of performance and quality, and whether tools like Cursor or local codex alternatives will fit well into my workflow. There’s a lot to figure out, but I’m focused on getting a stable, polished release out the door first.

Two-Line Takeaway

Key step forward

The most important forward progress step is the big refactor of AppState and Control Center, significantly reducing code complexity and improving stability while upgrading to newer AI models and optimizing performance based on system VRAM.

Significant barrier or risk

The most important barrier or risk is the unresolved bug where the app doesn’t abort cleanly during mid-transcription or summarization, alongside uncertainties about effectively communicating the app’s integration value and whether skipping beta and targeting GameDev communities will aid adoption.

Keynotes

Decisions & Observations

• Decided to skip the beta phase, removing beta feedback prompts and reworking contributor thank-you section; relying mostly on existing feedback for polishing.
• Upgraded to OpenAI models 4.0+, tested Mistral 7B and Qwen 3 models; implemented auto-optimization based on VRAM to balance quality and speed.

Issues & Next Steps

• Shifting marketing focus toward GameDev communities to leverage integration features; planning short video demos to showcase workflows.
• Needs to resolve a critical bug where the app does not abort cleanly if stopped mid-transcription or mid-summarization.
• Documentation on model performance and pricing remains sparse, making it difficult to confirm if current model choices are optimal.
• Considering use of Cursor subscription for code refactoring and repo analysis, but effectiveness is uncertain.
• Exploring feasibility of local codex-like tools for game dev or small teams inspired by Watchlight Studio’s offline approach; unclear if suitable options exist.
• Prioritizing polish and stability improvements ahead of May 15th release deadline.
• Open questions: how to fix abort issue during transcription/summarization, how to better communicate integration-focused value, whether skipping beta and targeting GameDev will aid adoption, and if current model setup balances performance and quality well.

Uncategorized Notes

• Completed major refactor of AppState and Control Center, breaking code into smaller files and reducing size; testing refactor on a separate branch to keep main codebase stable.

Generated with Nucleate by Watchlight Studio — distilled from full transcription.