▦ BETA · MADE IN BHARAT

India's first
community-powered
AI network.

200 million laptops sit idle across India every day. Sangam connects them into a peer-to-peer AI network that runs 70B-parameter models via swarm inference, lets you fine-tune private LoRA adapters on your own data, and opens a marketplace where you can sell what you've trained. And when you need raw firepower — the world's best frontier models (Claude Opus, GPT-4o, Gemini 2.5 Pro) are bought in bulk and resold at fair credit prices. No $50,000 GPU required. No data colonialism.

nodes online
contributors
tokens served
LoRAs trained
Why this matters now

India's AI future can't depend on three US companies.

200M+

Indian laptops sit idle most of the day. Sangam puts them to work — no new hardware needed.

22

Official Indian languages, most underserved by GPT/Claude/Gemini. Sangam is Indic-first by design.

₹0

No GPU rental to run a 70B model. Pay in credits you can earn by sharing your own laptop.

100%

Your training data stays on your machine. Federated learning means only LoRA adapters are uploaded.

Four tiers — one network

How Sangam works.

1

Lend your laptop — earn credits.

The Sangam node-client is a tiny Python program (~12 MB). Install it, paste your worker token, and let it run in the background. Your CPU/GPU becomes part of the swarm — your machine holds a few transformer layers of Llama 3 70B while 20 other laptops hold the rest. India's first Petals-style fork (BigScience architecture), built for Indian bandwidth.

⏱ Every minute online = 2 credits (GPU-equipped laptops earn 4×). 7 hours running → 840 credits → 30+ free model queries.

2

Spend credits on the world's best models.

Your credits unlock the entire Sangam model bazaar — both community-served open models (Llama 3 70B, Mixtral, Sarvam-1, Airavata) and frontier commercial models we buy at wholesale and resell (Claude Opus 4.7, Claude Sonnet 4.6, GPT-4o, Gemini 2.5 Pro, DeepSeek R1, Mistral Large, Grok 4). One login, one balance, every major LLM on the planet.

✦ Open Llama 70B = 25 credits / 1k output tokens. Claude Opus 4.7 = 240. GPT-4o = 60. Sarvam-1 (Hindi) = 4. Always 30-70% cheaper than buying direct.

3

Fine-tune your own LoRA — on your own laptop.

20 chartered accountants can co-train a LoRA on their ITR documents. 50 doctors on their OPD notes. 30 lawyers on contract drafts. Federated learning means the raw data never leaves any laptop — only the merged LoRA adapter (10-100 MB) is uploaded. Built on Flower + HuggingFace PEFT with India-friendly licensing.

⚒ Training cost = 50 credits base + 1 credit per MB × epochs. 25-credit completion bonus when the job finishes.

4

Marketplace — sell your LoRA, buy others.

"Tax-AI for CAs" — 200 credits. "Class-12 Physics tutor in Hindi" — 50. "Indian rent-agreement reviewer" — 150. This is the Hindi HuggingFace — but for Indian creators, with Indian payment rails. Sellers keep 90%; Sangam takes a flat 10% network fee. Sellers also earn royalties on every inference run against their LoRA.

₹ 10% network fee. No HuggingFace account needed — just Glixy login.

Architecture

20 laptops = 1 supercomputer.

Petals → India-fork.

A 70B model is split into ~80 transformer layers. Each laptop holds 3-4 layers. Your prompt travels through the swarm — every node computes only its slice; the full model is never reconstructed in one place. Token latency: 2-5 seconds (good enough for chat).

Inference works today. Training doesn't — that's what federated LoRA is for. This combination — peer-to-peer inference plus federated training plus a creator marketplace — has never been deployed India-first before.

Who is this for?

Sangam works for you if you're…

🧑‍💻

Developer / Indie hacker

Build on Llama 70B or Claude Opus without a $1,000 OpenAI bill. SDK-style API.

🎓

Student / Researcher

Earn GPU credits by running your laptop. Your thesis or side-project runs free.

🧾

CA / Doctor / Lawyer

Train a private LoRA on your domain — ITRs, contracts, case notes — without leaking data.

🏫

EdTech / Tutor

Build Hindi-medium subject tutors — Class 12 Physics, NEET Biology, UPSC prep.

🏪

Small business

Fine-tune on your products and FAQs → ship a WhatsApp customer-care bot. No per-call API fee.

🌍

Anyone in India

Sangam's backbone is Indian laptops. Every minute you contribute is a vote for sovereign AI.

500 free credits on signup.
Start in 5 minutes.

Create a Glixy account → open the Sangam dashboard → lend a laptop (optional) → chat with Llama 70B or Claude Opus. Done.

Bittensor-style token economics, but India-legal and RBI-compliant. No crypto — just in-network credits.