How you can use
Madhuram?
The API is open to everyone today. Finetuning and on-device builds are delivered as custom engagements, so get in touch and we will tailor them to your case.
Long-document understanding
A 16K context window takes a full paper, support thread, or meeting transcript in one pass, for summarising, extracting fields, and answering questions grounded in the source.
Build on the API
An OpenAI-compatible endpoint, so existing code works by changing a base URL and a key. Free while Madhuram-v0.6 is in preview, with streaming and the usual sampling controls.
Task-specific finetuning
At this size a model is cheap and quick to specialise. We finetune Madhuram on your labels for classification, extraction, or routing, and hand back a model that fits your task rather than a general one that approximates it.
Talk to us
On-device deployment
Around 100 MB quantised, fast enough for interactive use on a mid-range phone, with nothing leaving the device. We work with you on the build, quantisation, and integration for your hardware.
Talk to us