Took us a while to build this from scratch. She has some good ideas for building a action RPG game like Dragon Quest and I decided to give this a shot and she has been playing and testing the games. So far she's loving it. Hope you all enjoy.
Author here. Quick notes on how this works and where it's going:
Each video is built by a small Python pipeline — it pulls a notable HN gaming story, drafts a tight narration script, renders it with a neural TTS voice (running locally on Apple Silicon), and cuts it over footage and original motion graphics with ffmpeg. The site itself is a deliberately HN-flavored list with a simple upvote API.
It's early and very much a work in progress — right now there are a handful of episodes (Playdate, Tiny Wind, Battle for Wesnoth, an F-15 story, and a couple more). My goal is short, watchable mini-documentaries rather than auto-generated filler, so I'm iterating on pacing and writing quality more than volume.
Two things I'd genuinely like feedback on: (1) which HN gaming stories you'd most want to see covered, and (2) whether the narration lands as "documentary" or still feels robotic.
Footage sourcing is the part I'm most careful about, so I'm leaning more on original graphics over time.
Thanks for taking a look.
Bonsai: A Local Agentic AI Harness Built Around Small Models
Since last year, I've been teaching a course at UT Southwestern Medical Center on how to build Agentic AI systems and harnesses for specialized domains.
One thing I've noticed is that as companies like OpenAI, Google, and Anthropic continue raising API prices, the cost of running frontier models in the cloud keeps increasing. At the same time, many users are using ChatGPT the same way they used Google years ago: asking questions and looking up information. Most of these use cases simply don't justify paying for GPT-5.5, Opus 4.8, or other expensive flagship models.
That led me to explore a different idea: combining efficient local models with a purpose-built harness that provides tools, memory, and domain-specific skills.
Part of the reason I named this project Bonsai is that I had some interactions with Stanford's Prism Lab. The architecture follows an Agent + Skills + Memory design. Memory is implemented locally using embeddings and SQLite, allowing semantic retrieval through cosine similarity search. This helps compensate for the limited context windows of smaller local models.
I believe this approach can make small models much more capable than their parameter count would suggest.
Although Anthropic has never publicly disclosed the exact size of Claude Sonnet, my analysis suggests it is likely a Mixture-of-Experts (MoE) model with tens of billions of active parameters and hundreds of billions of total parameters.
The active parameters determine how much computation is used during inference, while the total parameters represent the model's stored knowledge. My hypothesis is that a dense thinking model with only tens of billions of parameters can still deliver strong performance if paired with effective harness engineering, specialized tools, memory, and retrieval systems.
If that hypothesis is correct, local models could satisfy the majority of everyday ChatGPT-style use cases without requiring expensive cloud inference.
As a first step, I'm releasing an experimental version of Bonsai.
Bonsai communicates directly with a local Google Chrome instance and provides a collection of browser-oriented tools that allow a local LLM to interact with the web in an agentic fashion. The default model is Google Gemma 4B, although Qwen models can also be used.
(One reason I chose Gemma as the default is that some government agencies and schools in Texas prohibit the use of Chinese open-source models.)
The left side shows the chat interface, while the right side displays the agent operating the browser in real time.
The harness includes many browser-specific tools, including JavaScript injection capabilities that allow the agent to locate page elements, inspect DOM structures, click buttons, fill forms, and perform other browser interactions.
Current features include:
Browser integration
VectorDB-based semantic memory for small-context local models
Custom browser-oriented skills and tools
Local embedding + SQLite memory system
Agentic web navigation
WebRTC-based communication layer (lower-level than MCP)
The current release was compiled for Windows and requires NVIDIA CUDA.
I've also added an Apple Silicon (M-series) Mac version to the same download directory.
The default model is a 4B thinking model because agent workflows benefit significantly from high token throughput. On my test system (Windows 11 + RTX 4090), Bonsai reaches roughly 140 tokens/sec. On an M4 Mac using Metal, I see around 50 tokens/sec.
I'm curious whether others think specialized harness engineering can make small local models practical for everyday AI workflows, rather than relying exclusively on increasingly large cloud-hosted models.
I'm working on a side project, ap-quiz.com, has officially launched. It is a mobile-focused high school AP exam practice app that brings together a massive AP question bank. When you're waiting for the bus or waiting for your food at a restaurant, you can easily open the site and practice questions anytime.
The platform uses a game-style system with different levels, scores, and trophies, allowing users to compete with each other to see who can achieve higher scores. For questions answered incorrectly, AI provides targeted analysis so users not only learn the correct answer, but also understand why they got the question wrong.
Currently, the website is not completely free. It costs $4.99 per month (currently discounted), which helps us purchase more question banks and add more AP exam categories.
I just added smaller airport. (for now I turned off airport overlay as default since it get a little clutterd but you can turn it on. or you can search the airport code in search bar)
