Launched this week
Overmind
Open-source platform for continuous learning
561 followers
Open-source platform for continuous learning
561 followers
Overmind is the model training platform for AI teams. Turn your production traces into specialised models you own, trained, benchmarked, and served.
Super happy to be launching Overmind today 🔥 want to get started? Paste this onboarding prompt into the coding agent of your choosing and let it work:
"Set me up on Overmind https://console.overmindlab.ai/setup.md"
@samsmallmind Bish bash bosh!
This is legitimately really cool! Agent training, evals, observations, etc is such a fragmented and clunky process and I loveee what Overmind has done to make so much simpler! 🚀
@parthkoshti thanks for the support Parth!! AI has its own interface, so we thought training AI needs one too - as easily accessible and usable :)
@parthkoshti Loveee to hear it! Clunky was the exact word in our heads when we started. Now it is one prompt in your coding agent and you are off. Would love for you to take it for a spin and tell us where it still feels clunky, we will fix it 🙌
@tessak22 Thanks Tessa! It's as easy as "Set me up on Overmind https://console.overmindlab.ai/setup.md"
Let us know how you get on!
@tessak22 Thank you Tessa! Can’t wait to hear what you think. If you have any agents doing repetitive GTM work like lead scoring or routing inbound, that is exactly where a small model trained on your own traces tends to shine. Happy to pair on it if you want a hand 🙌
this is the future!
Congtats on the launch guys
@mituhin Thank you Paul! We think so too 😄 Curious if you have any agents running at Flowmapp yet, would love to hear what you would train first 🙌
@mituhin thanks Paul! ❤️🔥
All the best!
@preetmishra Thank you Preet!! Means a lot 🙌
@preetmishra Appreciate it Preet! ❤️🔥
this could really changed the game 🔥
@princeajuzie Thanks Prince, really appreciate it! Curious if there is anything in Pastely you could see handing to a trained small model?
@princeajuzie That is the dream 🔥 Thank you Prince!
this is actually really cool.
@new_user_2790a57d4d Thank you Faizan! Means a lot 🙌 Are you running any agents at DocsAlot? Doc-heavy work like tagging and extracting fields tends to be a great fit for a small model trained on your own traces
@freyam7 yes I am. Not only that I was more thinking for stylic models, small specialized models that retail a particular style.
@new_user_2790a57d4d Ooh, style is a really fun one! It is actually a great fit, because style is exactly what lives in your own outputs and nowhere else. The tricky part is usually defining what "on style" means so you can measure it, and that is where the evals come in: you define it once, and every new model gets scored against your current one before you swap anything. What kind of style are you thinking, tone of voice for the docs or something more structural? 🙌
@new_user_2790a57d4d We think so too! Just had a look at DocsAlot (cool product btw) - docs automation could be trained, checking every PR for whether the docs needs an update is the same job over and over across every customer's repo. A small model trained on past decisions could likely do that for much less per call than what model you're using atm. Which model are you using?
@samsmallmind You're right. We use openAIs model with vercel AI SDK for our agents for synching the docs with other signals.
Its a hodge-podge of things, we've been meaning to cleanup. Are there free credits, to try overmind?
@new_user_2790a57d4d Love that, Faizan! Yes, there are $50 in credits to get you started, and the whole platform is open source if you would rather run it yourself 🙌 A "hodge-podge" agent is honestly the best place to start, because the traces show you exactly which steps are worth training a small model for. Shout if you want a hand getting set up!