Gargi-M1 is a 110M-parameter research preview with a 512-token context. Its output is often wrong or nonsensical and is generated as-is — Gargi accepts no responsibility for how it is used or interpreted. Conversations are stored to improve the model — see privacy.