Overmind (@OvermindLab) header
Overmind (@OvermindLab) profile picture

The open source training platform for continuously improving agents and models

London & San Francisco
Joined January 2026
Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)Photo by Overmind (@OvermindLab)
Overmind turns anyone into an AI lab. • Builds a context graph from code and traces • Curates training and eval datasets • Evaluates prompts and models • Trains smaller, specialized models github.com/overmind-core/ove…
88
38
54
320
758,765
18 models trained for one job: tool calling. The results? LFM3 (1.7B) 2 points under Gemini 3.1 Pro but 3 times faster. Two trained 27B Qwen models scored higher than the frontier. Training a model for a specific job is now an accessible, routine part of building software.
10
2
1
18
383
Overmind retweeted
Overmind added @liquidai's LFM3 model last week. Below, the 1.7B model was trained to do one thing: pick the right tool for a request and fill in the details. On 1,000 test requests: • LFM3 scored 81.8% • Gemini 3.1 Pro scored 83.9% • LFM3 answered 3 times faster
3
3
9
256
Overmind added @liquidai's LFM3 model last week. Below, the 1.7B model was trained to do one thing: pick the right tool for a request and fill in the details. On 1,000 test requests: • LFM3 scored 81.8% • Gemini 3.1 Pro scored 83.9% • LFM3 answered 3 times faster
3
3
9
256
LFM3 was one of 18 models Overmind trained and tested the same way. Each dot is one of them, placed by size and score. Train one for your own job: docs.overmindlab.ai/tutorial…
1
6
101
Overmind retweeted
the 383 vs 123 is one $200 live test, and @notmiramurati already said the live a/b tests were messy the more solid number is on their held-out set: the same @deepseek_ai v4.1 flash went 64% → 77% after training on traces, above gemini 4 argon at 70% and grok 4.7 at 64%. they split train/val/held-out and published the traces that's the pattern. pick an open model, train it on your task, score it against what you already run, take the weights this is the loop we built @OvermindLab for, minus the custom simulator. if you already have prod traces you can start there
Meet Yvligo, our open-weights model for running X ads. (3x better than Grok 4.7) At loglog.ai, we built an X simulation and post-trained DeepSeek V4.1 Flash, outperforming frontier models Live results: Yvligo: 383 relevant website visits for $200 Grok: 123 visits
3
6
205
A step an agent repeats can run on a small model trained on its own past runs. Turn your company into an AI lab. Available via your coding agent of choice.
One fine-tuned 2B model topped Airbnb's chart, above every frontier model they tested. That's one classifier. Agents make calls like that on every run: • model routing • tool calling • input parsing Imagine each of those as a small model trained on your own traces, and retrained as new traces come in. Every run makes the next version better, at every step. A world where each non-deterministic step continuously learns and improves... ♺
1
7
226
Train your model on Overmind, then serve it there. Dedicated GPUs for the models you train, with the option to scale to zero when idle. An addition to our serverless inference.
Yesterday a customer asked us for a feature we hadn't shipped yet. So like the cracked founders we are we pulled an all nighter and got it deployed! @OvermindLab now offers dedicated inference for your models.
2
2
12
365
yep. every deployment gets its own pinned model id, and a pinned id always hits that exact deployment for prod, point your app at the capability alias instead. it only moves when you activate a new model, so you can run old vs new on the same eval set first, then switch
1
44
How do you define "good" for an agent? How do evals stay honest when the agent changes? @freymehta answered questions all day. Thank you to everyone who tried Overmind and asked the hard questions. Onwards.
spent launch day in the @ProductHunt replies with builders and founders actually trying overmind 🚀 skeptics asking how you even define “good” for an agent. a founder going four rounds deep on how the training loop works. builders poking at how evals stay honest when the agent changes underneath them. a few of those threads are already shaping what we build next we landed #2 product of the day, just shy of the top spot, which was a lovely bonus on an already great day. by the evening we were all glued to the leaderboard and i was back in 2016, refreshing my youtube channel until it hit 1,000 subs!! huge shout out to @samsmallmind for leading the whole launch, and to claude for turning the day into this little video (it would not stop adding pixels!!) 👇
1
2
8
254
Overmind's documentation now works for you and your coding agent. Take a prompt from the docs and train a model with it in Claude Code, Codex or Cursor. Every tutorial and how-to guide has one. Turning any company into an AI lab, made easy. It starts with a prompt.
4
2
1
15
3,476
What does it mean to break Big Tech's grip on enterprise AI? • When a company rents an AI model, rivals can rent the same one. • The edge comes from training AI on how it's used. AI labs do this. Now a company can too. Overmind turns any company into an AI lab.
1
1
1
9
257
Overmind's CEO spent years at MI5 & MI6 relying on AI that was always built by someone else. "We simply couldn't trust AI to make life-or-death decisions." So Overmind lets a company train AI on its own data, fitting that business and staying in its hands. Any team can do it.
2
2
2
13
620
Thanks to @iamVictorDey and @Forbes for the write-up. Full piece here: forbes.com/sites/victordey/2…
4
84
Overmind retweeted
Overmind is live on Product Hunt! 🚀 Some big companies have trained their own models this year. That took an ML team. Most companies don't have one. They do have coding agents though. So Overmind lets those agents do the ML team's work. Live: producthunt.com/posts/overmi…
34
4
3
53
19,412