e
electr0
3 models is a great start but the ability for users to download and use additional models locally would be ideal. Either via a huggingface integrated search or additional curated writing models (Gemma 3 270m is great and lightweight)
Another feature for paid users would be writing styles. This would require the user to upload their work or select it from previous project files and Amber would create a lightweight LoRA in their style that they could select within the AI assistant input to to the right of "Add Context" i.e. "Add Style"
Example Workflow in App:
  • User uploads personal writing documents or selects previous writings.
  • App prepares dataset and initiates LoRA fine-tuning job in the cloud.
  • LoRA adapter is trained and saved under user profile locally.
  • User selects their adapter as "writing style" for chat generation.
  • Generation backend loads the base model plus selected adapter (writing style) for inference, ensuring outputs match user’s style.