Taro

Today I’ve added a skill for researching a topic. I based it off the research skill on Daniel Miessler’s Personal AI Infrastructure project, but modified it to be compatible with OpenCode, use less tokens and have a response format that I like better. The research skill works by first generating different research angles. After that, a subagent is spawned in parallel for each research angle. Finally, the findings from each agent are combined into a research report. The result is the prompt “research everything we know about consciousness and different theories and philosophies” generating this result:
(full answer on https://gist.github.com/Mikuel210/a8589b2f922675819077e517cca3dac7)

Today I’ve added a skill to get context on the user from a file when needed (e.g. get the projects they’re working on)

Today I’ve added a tool for getting the current time. I’ve also updated the system prompt slightly to tweak the personality.

Taro is my attempt at building the coolest AI project ever that makes you feel like we’re living in the future. I really think AI has got a huge potential, but right now it’s hard to integrate it into our daily lives to automate our day to day tasks. My goal is to build a system that understands your context and intentions to solve problems and automate tasks.

Taro is being built on top of OpenCode, which provides the foundations for agentic AI, including tools and skills. However, I plan on extending its capabilities by making the AI be able to have initiative when certain events are triggered. I also plan on making a cross-platform client for using it from anywhere.

The main component of the system is skills. Skills load context on a specific domain, and they include workflows, which instruct the AI on how to perform a task achieve a specific result. This is huge because you can create deep context on how you want the AI to do things and have it only load the context when needed.

For now I’ve been experimenting with OpenCode and creating my first tools and skills. I’ve coded a tool for getting YouTube transcripts and another for getting transcripts from images and documents. For now, I’m going to work on making more tools and skills to solve not only problems that I have but for general tasks as well that I think could help for other people.