5 ms·

If you're using Cursor (as I do), but find it tends to struggle once your project reaches a certain size, and you're ok with something CLI-based, you might find

If you're using Cursor (as I do), but find it tends to struggle once your project reaches a certain size, and you're ok with something CLI-based, you might find the open source CLI coding agent I've built interesting: https://github.com/plandex-ai/plandex https://github.com/plandex-ai/plandex - It can handle up to 2M tokens of context directly, and can index/work with/chat with projects up to 20M tokens (1M+ lines). Here's an example of chatting with with SQLite codebase to learn about how transactions are implemented: https://plandex.ai/_next/static/media/plandex-sqlite.0ee6cb28.gif https://plandex.ai/_next/static/media/plandex-sqlite.0ee6cb2... - All changes are committed to a version-controlled sandbox by default, preventing the problem of stray changes that you don't notice being left behind in the project. - Being terminal-based allows for more seamless and powerful execution control and automated debugging. Here's an example of automatically debugging a browser app (via redirection of console logs/errors to the terminal): https://www.youtube.com/watch?v=g-_76U_nK0Y&embeds_referring_euri=https%3A%2F%2Fplandex.ai%2F https://www.youtube.com/watch?v=g-_76U_nK0Y&embeds_referring...

colesantiago 1y ago
I see too much self promotion over this tool all over HN and it is not anywhere near the same as Cursor? I don't see any similarities other than it uses AI and that is about it.
- danenania 1y ago
  If you don't see any similarities, I'm not really sure what to say. I built it specifically to improve the workflow of what has come to be labelled 'vibe coding', and I believe it's more effective for this purpose than Cursor... though I find that Cursor is excellent for IDE auto-completion. It's MIT-licensed and can be used for free, so yes I do mention it when relevant conversations come up, because I think can be useful to people, and I think that is well within the spirit of HN, which is supposed to be a maker community. Since I'm an HN addict, I read/post a lot, and so I notice when these topics come up.
kristopolous 1y ago
I've used it. Honestly it's the same place as goose and aider ... it tries to DIY its own solutions too often as opposed to using existing robust libraries, writes inflexible code where simple piped commands would do, and basically Katamari Damacies the problem into a complicated sprawl that doesn't work.
- danenania 1y ago
  Thanks, I appreciate the feedback. I think the reason you tend to see these same kinds of problems across tools is that you are running up against the reasoning limitations of the underlying models. I see my goal as trying to push that horizon out as far possible, and I think Plandex pushes it significantly farther than most other tools, but you do of course still run into those limitations. That said, I think even setting aside improvements to the underlying models, there's a lot of potential to improve on these issues. I think they're all basically addressable at current capability levels, though they are difficult problems. It's what I'm personally most excited to work on.
  - kristopolous 1y ago
    Honestly, I think it simply becomes the wrong modality at a certain level of engagement with any problem. When human developing a project successfully you engage in different collective behaviors throughout the process. That's why the scrum/agile stuff that tries to normalize it doesn't actually work very well. On computers, relationships are workflows and you need to do a dance of fluid relationships to use AI effectively throughout the execution process - otherwise that's why you either abandon it after you get to some point or you feel like you're just wasting time.
    - danenania 1y ago
      You're right. I think the solution essentially boils down to modeling those other modalities and incorporating them into the agent.

colesantiago 1y ago

I see too much self promotion over this tool all over HN and it is not anywhere near the same as Cursor? I don't see any similarities other than it uses AI and that is about it.

danenania 1y ago
If you don't see any similarities, I'm not really sure what to say. I built it specifically to improve the workflow of what has come to be labelled 'vibe coding', and I believe it's more effective for this purpose than Cursor... though I find that Cursor is excellent for IDE auto-completion. It's MIT-licensed and can be used for free, so yes I do mention it when relevant conversations come up, because I think can be useful to people, and I think that is well within the spirit of HN, which is supposed to be a maker community. Since I'm an HN addict, I read/post a lot, and so I notice when these topics come up.

kristopolous 1y ago

I've used it. Honestly it's the same place as goose and aider ... it tries to DIY its own solutions too often as opposed to using existing robust libraries, writes inflexible code where simple piped commands would do, and basically Katamari Damacies the problem into a complicated sprawl that doesn't work.

danenania 1y ago
Thanks, I appreciate the feedback. I think the reason you tend to see these same kinds of problems across tools is that you are running up against the reasoning limitations of the underlying models. I see my goal as trying to push that horizon out as far possible, and I think Plandex pushes it significantly farther than most other tools, but you do of course still run into those limitations. That said, I think even setting aside improvements to the underlying models, there's a lot of potential to improve on these issues. I think they're all basically addressable at current capability levels, though they are difficult problems. It's what I'm personally most excited to work on.
- kristopolous 1y ago
  Honestly, I think it simply becomes the wrong modality at a certain level of engagement with any problem. When human developing a project successfully you engage in different collective behaviors throughout the process. That's why the scrum/agile stuff that tries to normalize it doesn't actually work very well. On computers, relationships are workflows and you need to do a dance of fluid relationships to use AI effectively throughout the execution process - otherwise that's why you either abandon it after you get to some point or you feel like you're just wasting time.
  - danenania 1y ago
    You're right. I think the solution essentially boils down to modeling those other modalities and incorporating them into the agent.