Models
We're building small, utility-scoped models for specific tasks — fine-tuned from open foundations like Gemma 4.
What we're working toward
Our models will be focused and practical: domain-specialized adapters trained with LoRA/QLoRA, optimized for specific tasks rather than general capability.
Each model release will include full documentation — training methodology, benchmark results against standard evaluations, intended use cases, limitations, and weights available for download.
No models released yet. Follow our research for progress updates.
Read about how we approach model training in our research, or learn more about the team.