Launched this week
Overmind
Open-source platform for continuous learning
571 followers
Open-source platform for continuous learning
571 followers
Overmind is the model training platform for AI teams. Turn your production traces into specialised models you own, trained, benchmarked, and served.
Free Options
Launch Team / Built With
Framer AI AgentsDesign and publish professional sites with AI
Promoted
multiple years of AI experience told me turn any data into training data is very hard tbh. not data format itself, how to define the label, how to collect enough signal, which data is good to use through hundreads of thousands of table, also you need to have a really deep understanding for the business domain as well. Also curious how you handle entire mlops, how to minotor the drifting and model retraining, and infrerence cost
@gavin_fgg0815 Totally agree Gavin, and honestly this is the exact problem we spent most of our time on. We do not try to turn "any data" into training data. We start from the agent's own production traces, so the inputs are already the real task. The label problem becomes defining what "good" means for each capability, and that is where the domain knowledge lives, so the team owns it as evals. The evals are also what catch the "no signal" case. We had a run where every model, big and small, sat at ~67% on a yes/no task, and Overmind showed that in the numbers instead of letting us ship a coin flip.
On the ops side, traces keep getting scored on live traffic, so you see drift as it happens. New traffic becomes the next versioned dataset, and any new model has to beat your current one (objectively) on the same frozen set before it goes anywhere. Inference sits behind one OpenAI-compatible API, so you can run your own small model and a frontier one side by side and see the cost difference yourself (similar to A/B testing).
Would genuinely love your take after poking at it, this is the kind of feedback that makes it better 🙌 @pritamsonihsr can go much deeper on the eval side too
@gavin_fgg0815 Freyam covered most of the bits around datasets,
to monitor agent's signals and to solve drifting, we've binding evals to live traces, so as your agent evolves your evals metrics will get updated from code, and this helps us solve drifting problems.
and these evals metrics are used with a decision model to score the entire conversation to build signals, which can then be used as feedback for continuous improvements of agents.
Heyyy PH, Frey here, founding engineer at Overmind.
When most teams hear "build your own AI", they picture a research lab, a GPU budget and a year of hiring. We think it should feel more like flipping a switch. So we built one open-source platform that watches every agent you have in production, learns from your own data, trains models made for those agents and keeps making them better well after launch. All of it runs on your own infrastructure, so your data never leaves your hands. The traffic you see today turns into tomorrow's training set, and your agent gets a little better every day.
We kept one test in our heads the whole time: can someone who has never fine-tuned a model go from one prompt in their coding agent to their own model in production? Making the answer yes is what the whole team spent the last few months on.
It is all open source, so if something looks off, open an issue or come find me in the comments. Happy to go as deep as you want on the technical side!
Toodaloo 👋
@freyam7 toodaloo!
@freyam7 looks amazing. Great work team and congrats on big launch momentum already.
@hissingkitty Thanks!! The team can't wait to see everybody start "creating" AI rather than just consuming it!
@freyam7 @hissingkitty thanks Christine! just getting started :)
@freyam7 Congrats on your launch today. Weldon to you and the team.
@firdausi_adamu1 Thanks for the kind words, Firdausi! 🙌
Happy launch team! Sounds like interesting project, I would like to check it out. But why not adding more information in the description field? Most people here don't really check the screenshots or videos ;)
@davitausberlin but our branding is so cool 🥺 - you're right and thanks for looking out! The main comment doesn't seem to be pinned, will relay it here:
Overmind is the open-source platform for AI agent training. What we're working towards is giving every product its own AI lab. 🚀
Most AI products run on the same few frontier models. A frontier lab works differently, collecting data, building evals, training, testing. Ship and repeat.
To get started: Connect our MCP, and paste one prompt into Cursor, Claude Code, Codex or OpenCode.
What you get:
🔸 Tracing: OpenTelemetry traces, scored as they arrive and matched to the part of your agent that produced them
🔸 Datasets: production traces turned into versioned eval and training sets
🔸 Evals: you define what "good" means for each capability, and Overmind measures it on live traffic
🔸 Optimisers: prompt, tool and code experiments run in your repo, and the winner comes back as a git diff
🔸 Fine-tuning: small specialised models trained on your data and benchmarked against the model you run today, with weights you can download and export
🔸 Inference: your own models and frontier models behind one OpenAI-compatible API
Why are we doing this? 💭
Today, millions of developers can build with AI, but creating and improving the models behind it still requires expertise concentrated in relatively few teams.
We believe that gap will close. Every application, business, device and individual has different needs. Meeting those needs will require more people to be able to create, test and run specialized models.
Computing became widespread as processors became cheap enough to build into almost anything, and development tools made them useful to far more people. We believe intelligence needs a similar expansion in who can build it and where it can run.
Overmind is our contribution to that: putting the work of an AI lab into tools any developer can use, so they can create, own and improve intelligence for their own tasks.
Our mission is to make intelligence as easy to create as it is to consume.
Hey team, this sounds crazy, I always had this mental friction in my head that training model is a tedious task, and needs some decent technical bandwidth. But this sounds like it'll remove that friction.
@priyansh_agrawal2 Thanks Priyansh! That friction is what we built Overmind to remove - your coding agent handles the data prep and training for you so don't need ML experience to get a trained, smaller (and very often better) model out. Anything you can think in Shadow where Overmind would help?
Hey @priyansh_agrawal2 ! That friction is exactly what we were chasing! Your coding agent does the ML team’s job: it turns your traces into evals, trains a small model and benchmarks it against what you run today. One prompt to start: "Set me up on Overmind https://console.overmindlab.ai/setup.md". Curious what you would train first for Shadow 🙌
Hey Product Hunt 👋
I'm Tyler, co-founder and CEO of Overmind.
Overmind is the open-source platform for AI agent training. What we're working towards is giving every product its own AI lab. 🚀
Most AI products run on the same few frontier models. A frontier lab works differently, collecting data, building evals, training, testing. Ship and repeat.
To get started: Connect our MCP, and paste one prompt into Cursor, Claude Code, Codex or OpenCode.
What you get:
🔸 Tracing: OpenTelemetry traces, scored as they arrive and matched to the part of your agent that produced them
🔸 Datasets: production traces turned into versioned eval and training sets
🔸 Evals: you define what "good" means for each capability, and Overmind measures it on live traffic
🔸 Optimisers: prompt, tool and code experiments run in your repo, and the winner comes back as a git diff
🔸 Fine-tuning: small specialised models trained on your data and benchmarked against the model you run today, with weights you can download and export
🔸 Inference: your own models and frontier models behind one OpenAI-compatible API
Why are we doing this? 💭
Today, millions of developers can build with AI, but creating and improving the models behind it still requires expertise concentrated in relatively few teams.
We believe that gap will close. Every application, business, device and individual has different needs. Meeting those needs will require more people to be able to create, test and run specialized models.
Computing became widespread as processors became cheap enough to build into almost anything, and development tools made them useful to far more people. We believe intelligence needs a similar expansion in who can build it and where it can run.
Overmind is our contribution to that: putting the work of an AI lab into tools any developer can use, so they can create, own and improve intelligence for their own tasks.
Our mission is to make intelligence as easy to create as it is to consume.
The team is here all day. What would you train first?
The humans at Overmind ❤️ 🌎
@kashish_tripathi Great question Kashish! It works best anywhere an AI agent does the same kind of job over and over: sorting support tickets, routing requests, scoring leads, pulling fields out of documents, picking the right tool. Overmind learns from how your agent already handles those jobs, trains a small model just for that task, and shows you side by side whether it beats what you run today. Are you building with agents at the moment? Happy to tell you if it is a good fit 🙌
Quick question: What are some use cases where this excels at?
@alchemist_studio Great question André! It shines wherever your agent makes the same kind of call over and over: routing, classification, picking the right tool, extracting fields. Those are exactly the jobs where a small model trained on your own traces can match or beat a frontier one, and you see the before/after numbers before you swap anything. What does Opengeni’s agent spend most of its calls on? Happy to tell you if it is a good fit 🙌
@alchemist_studio Jobs your agent does over and over - an example being sorting bank customer messages into 77 categories. We trained a 0.8B Qwen model on the agent's past answers and accuracy rose from 37% to 84%. The same approach works for any job your agent repeats, from simple ones like tagging leads, flagging risky payments, to more complex jobs like reviewing contracts.
Basically, if your agent does a job often enough and you (or we, via our evals) can tell a good answer from a bad one, you can train a model on it.
@samsmallmind Right. So thats the general use case. Bank support is the outcome use case. What other high value use cases are perfect for this? Like what kind of customers is this in high demand for? Other than bank chatbot?
@alchemist_studio The pattern we see is any team where an agent makes the same high-volume decision thousands of times a day and every call to a frontier model adds up: support triage and routing, lead scoring for sales teams, flagging risky payments in fintech, first-pass contract review in legal (to name a few). Regulated teams care a lot too, because a small model you own and run on your own infrastructure is far easier to get past compliance than sending everything to a third party API. Which bucket is Opengeni closest to? 🙌
@freyam7 Gotcha! What does the training pipeline look like? I would imagine the businesses that needs this are technically inclined. In the sense that they rather favor no fuzz UX ergonomics. They have oceans of data. But would rather just connect it and get a working small model out on the other end, that lands in their agentic product. and probably continuously as well, both input and output?