Building the AI co-scientist: a desktop agent, not a chatbot
Convai Innovations is an AI technology company based in Kerala, India. Scientific knowledge now doubles roughly every two months, faster than any researcher can read. We built Nadhi to close that gap: a desktop multi-agent co-scientist with system-wide file access. You pick the approach, RL, maths, or code, and a Planner, Critic, Director, Synthesizer, and sub-agents brainstorm to a final verdict, download 100+ open-access papers from arXiv, PMC, bioRxiv and OpenAlex, reads and cross-references files anywhere on your PC, runs the experiment, and writes results back into your manuscripts in the right format with a .bak undo. You stay the scientist who decides.
Our Vision
A world where every scientist has a tireless co-scientist. One that reads every relevant paper, connects findings across fields, and proposes grounded next steps, while the researcher keeps the judgement and the credit.
Our Mission
To raise the clock speed of research by putting a literature-scale co-scientist (deep research, cross-document synthesis, and grounded drafting) into a single desktop app any researcher can install and use within minutes.
Why we built Nadhi
Researchers, clinical scientists, and professors are drowning in literature. Findings that matter to a problem are scattered across PubMed, arXiv, bioRxiv, Nature and Cell, often in fields no single person can track. A browser chatbot answers from memory; it can't go and read the actual papers, and it can't reason across your own datasets sitting in a project folder.
Generic autonomous-agent frameworks promise the world but demand risky system permissions, complex setup, and have no grounding, so they confidently invent citations. In science, an unverifiable answer is worse than no answer.
We asked: what if a scientist had a co-scientist that actually fetches and reads hundreds of papers, synthesises them against the scientist's own work, and drafts hypotheses and reviews where every claim is traceable to a source it downloaded?
That's Nadhi: a desktop research agent paired with a frontier cloud model. It runs next to your files; only the prompt and relevant excerpts go to the model for inference, and no user data is collected or stored. Every output is grounded in sources you now hold on disk, and you remain the scientist who interrogates, validates, and decides. (Fully air-gapped, on-device inference for institutions is in development.)
Technology Stack
Local Multimodal Agentic LLM
State-of-the-art open model, optimized for on-device inference via LiteRT
Native Execution
Agent runs entirely via native TUI and tool calls securely on your machine
Electron Desktop
Native Windows application with hardware-encrypted license management
RAG Memory
Semantic vector search across all project documents using local embeddings
Multi-Gateway
Telegram, WhatsApp, and Email gateways for seamless communication
BYOM Support
Swap in any open-source model directly into the native runtime
Our Founders
Nandakishor M
Founder & CEO
AI/ML Researcher with secure agentic applications focus
Former Principal Investigator, IIT Palakkad Technology iHub Foundation
Dr. Anjali M
MBBS, MD
Co-Founder & Chief Research Officer
MD in Community and Family Medicine from AIIMS Rishikesh
Medical logic design & clinical validation of Ai4Cardio multimodal AI
Convai Innovations Pvt. Ltd.
Kerala Startup Mission
ASAP Community Skill Park
Kasaragod, Kerala, India
Incubated at Kerala Startup Mission. Building privacy-first agentic AI for Indian and global markets.
Licensed under AGPL-3.0 for the open-source core. Commercial desktop license available for research institutions and clinical labs.