{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreiasjypoklzcuxrm34rxnptpcwmssoghbaj2t7yfmdgsjctrtaqtbq",
"uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mmuwdkjujdo2"
},
"path": "/t/built-a-browser-based-rl-training-platform-with-huggingface-model-sharing-feedback-welcome/176276#post_1",
"publishedAt": "2026-05-28T00:04:43.000Z",
"site": "https://discuss.huggingface.co",
"tags": [
"agenlus.com",
"Launching on Product Hunt soon"
],
"textContent": "Hey HuggingFace community\n\nI’ve been working on Agenlus — a platform where you can train RL agents directly in your browser (no install, WebGPU-accelerated) and share your trained models to HuggingFace.\n\nThe idea is simple: train an agent on CartPole, MountainCar, or battle environments, then push your model straight to HF so others can load, compare, and build on it.\n\nRight now the flow is:\n\n 1. Train in browser → 2. Push to HuggingFace → 3. Others can load and battle your agent on the leaderboard\n\n\n\nThe HF integration was the piece I was most excited about — RL models have always been weirdly hard to share in a standardized way. HuggingFace already solved this for NLP and CV, so it felt natural to plug into that ecosystem for RL.\n\nWould love feedback from this community especially — are there model formats or metadata standards I should be following to make the models more useful/interoperable?\n\nagenlus.com Launching on Product Hunt soon",
"title": "Built a browser-based RL training platform with HuggingFace model sharing — feedback welcome"
}