Lipi: the open-source agentic scribe

Introducing Lipi: an open-source scribe for every conversation

Today, we are proud to open-source Lipi, an agentic scribe that listens to a meeting and turns it into clear, structured notes. Lipi (earlier known as Vaarta) works for any meeting, in any domain, and any organisation or individual can run it on their own infrastructure.

Lipi is built entirely on open building blocks: Eka Care's open Parrotlet speech model, the Echo agent SDK, and the newly published Scribe Exchange Protocol. Together, they give individuals, developers and institutions everything they need to bring an AI scribe into their own workflows.

From the clinic to every meeting room

At Eka Care, we built EkaScribe, an AI scribe that thousands of doctors across India use to turn consultations into clinical notes. Years of listening to doctors and patients, in many languages and accents, taught us one thing: the hard part of a good scribe is capturing what was said accurately, in real rooms, with real noise, and turning it into a useful note that people can act on.

We took that experience and built a scribe that is not tied to healthcare. Government meetings were the first use beyond the clinic. The same harness now extends to any domain, from review meetings and field visits to classrooms and boardrooms.

Lipi already powers multiple deployments for public health and meetings such as NHA VTT and Vaarta.

What Lipi does

Lipi sits in the meeting, listens, and hands back notes in the format each team needs.

• Clear audio, even in busy rooms. Improved audio capture and enhancement mean recordings from crowded halls and everyday microphones come out clean enough to transcribe well.
• Built for how India speaks. Transcription handles Indian English, Indian languages and the natural mix of the two.
• Works where the network doesn't. Lipi is built for jittery connections, low bandwidth and small laptops running many apps at once. It does not assume ideal conditions.
• Notes that fit the work. Templates turn a conversation into minutes, action items or any structured format a team defines.
• Grounded in your context. Teams can connect their own knowledge base, so notes use the right names, terms and references.
• Yours to run. Lipi can be self-hosted, keeping recordings and notes inside your own infrastructure, with secure sign-in through OIDC.

Open at every layer

We believe public-good AI should be open all the way down. Lipi itself is open source: the scribe application and the agentic harness that runs it. Underneath, it is built from three open pieces, each usable on its own.

• Parrotlet: Eka Care's open-source speech recognition models, built for Indian speech across English, Hindi, Marathi, Kannada and Telugu, including code-mixed conversations. Open-source language models then turn the transcript into structured notes.
• Echo SDK: Our open-source toolkit for building AI agents. It works with the model of your choice and powers Lipi's agentic workflow.
• Scribe Exchange Protocol: An open, vendor-neutral standard for how apps talk to any scribe service, in any field. Lipi is its reference implementation, so tools built for one compliant scribe work with another.

Build with us

Lipi is free to use, adapt and deploy. We invite developers, governments, institutions and startups to run it, build templates for their own domains, and help shape the Scribe Exchange Protocol with us.

Every meeting holds decisions worth keeping. With Lipi, capturing them is now open to everyone.

• Lipi on GitHub
• Parrotlet models on Hugging Face
• Echo SDK
• Scribe Exchange Protocol
• Eka Care open-source projects

And if you are a doctor, the scribe that started it all, EkaScribe, is at ekascribe.ai.