I'm not familiar with either since the inspiration for this came from other places, however, I've checked out both to get a better idea of what they do.
Firstly, Strands seems like a fantastic solution.
From what it seems it's a "hands-on" harness, which is targeting people with a know-how and a knack for tinkering.
VB Harness is not that, in-fact as time goes on, we're trying to make it as "hands-off" as possible & targeted towards the indie audience that are not too techie, but still want to get their hands dirty.
So you sacrifice a bit of control, in exchange for simplicity: i.e -
In Strands, you do need a bit of setting up for evals:
While yes you can extend them, it isn't the intention.
Another differentiating aspect is that our harness performs best with web-development work, and is tailored around AstroJS.
I'm not going to get into details, and will keep it brief. There are 3 reasons for Astro:
1 - HTMX talk on building on web standards, Astro builds on web standards.
2 - One of the easiest frameworks to use, apart from raw HTML/CSS
3 - We have an Astro fine-tune kit, with a sample dataset. I believe we were one of the first orgs to fine-tune a model specifically for Astro (atleast where we're based in, Sharjah). Here is the repo:
With everything said, the harness begins to really shine above everything else we've used (apart from SoTA cloud models/harnesses), for those particular reasons:
1 - It's simple
2 - It's easy to use, "batteries-included"
3 - It focuses on 1 domain, and excels in it because it specifically doesn't tailor to "everything" like Strands.
Hmmmm, what does this provide over what I'm doing atm which is Strands and GoalLoop?
I'm not familiar with either since the inspiration for this came from other places, however, I've checked out both to get a better idea of what they do.
Firstly, Strands seems like a fantastic solution.
From what it seems it's a "hands-on" harness, which is targeting people with a know-how and a knack for tinkering.
VB Harness is not that, in-fact as time goes on, we're trying to make it as "hands-off" as possible & targeted towards the indie audience that are not too techie, but still want to get their hands dirty.
So you sacrifice a bit of control, in exchange for simplicity: i.e -
In Strands, you do need a bit of setting up for evals:
https://strandsagents.com/docs/user-guide/evals-sdk/eval-sop...
In VB Harness, evals are provided out of the box, with 3 modes available (careful, balanced, quick):
https://vb-harness.vandbruno.net/start/how-it-works/
While yes you can extend them, it isn't the intention.
Another differentiating aspect is that our harness performs best with web-development work, and is tailored around AstroJS.
I'm not going to get into details, and will keep it brief. There are 3 reasons for Astro:
1 - HTMX talk on building on web standards, Astro builds on web standards.
2 - One of the easiest frameworks to use, apart from raw HTML/CSS
3 - We have an Astro fine-tune kit, with a sample dataset. I believe we were one of the first orgs to fine-tune a model specifically for Astro (atleast where we're based in, Sharjah). Here is the repo:
https://github.com/V-and-Bruno/astro-7-finetune-kit
With everything said, the harness begins to really shine above everything else we've used (apart from SoTA cloud models/harnesses), for those particular reasons:
1 - It's simple
2 - It's easy to use, "batteries-included"
3 - It focuses on 1 domain, and excels in it because it specifically doesn't tailor to "everything" like Strands.
[flagged]