Nihal Shetty
@nihalshetty0
Followers
136
Following
682
Media
100
Statuses
289
Building MVPs to launch your idea in weeks not months. Got an idea? DM.
Joined April 2021
RT @nihalshetty0: using @OpenCode_AI you can spin up a remote coding agent that has access to your codebase and can work on multiple git br….
0
1
0
i like that @cursor_ai gives primitives (like ctrl+k, ctrl+I) which makes the whole thing unopinionated. but i wouldn't mind some higher level workflows like create prds, create execution doc, etc. anything that would make ai coding less magic and more predictable.
0
0
0
using @OpenCode_AI you can spin up a remote coding agent that has access to your codebase and can work on multiple git branches. just like what openai and cursor does. this is seriously existing stuff!!.
opencode . ai – the coding agent with some name collisions that happens to have a typescript backend and a go frontend, is more than just a tui coding agent. it has a hackable api. Two examples 👇.
0
1
3
would you rather spend time nitpicking the perfect database or figure out what would make your users to tell their friends about it?.
1
0
2
RT @nihalshetty0: working on dropping this line😬:. "⚠️This is a work in progress and experimental.". 25 stars already for an oss that start….
0
1
0
working on dropping this line😬:. "⚠️This is a work in progress and experimental.". 25 stars already for an oss that started 4 days ago
0
1
5
RT @thdxr: opencode is hackable and embeddable. we need to document how to do this better but someone was able to build a web frontend all….
0
10
0
RT @nihalshetty0: update on building multi-modal ai tutor. crazy results!! it's generating animated graphics👀 0:17. switch on speaker🔊👇 htt….
0
1
0
both of these are god send:. - gitingest[.]com when i need to dump entire repos to llm, to understand them fast. helps me get inspo on code implementation. grep[.]app when the docs suck and i need real examples. what other tools should i be using?.
1
0
2
claude is the only model that can handle images in tool responses. other models just break if you do this.
0
0
0
hacking w/ opencode. they have this headless mode that you can use via api. the docs are not still clear though
0
0
4
i opened chatgpt .com, and firefox prompted this. what's firefox cooking?
1
0
7
RT @nihalshetty0: an actually useful SWE benchmark would be a LLM running as a coding agent to build tools/abstractions for common good. >….
0
1
0
an actually useful SWE benchmark would be a LLM running as a coding agent to build tools/abstractions for common good. > "build shadcn but with base ui". you cant hype up LLMs with this. It actually has to deliver.
0
1
5