Skip to content

Desearch-ai/subnet-22

Repository files navigation

Desearch

Subnet 22 on Bittensor

License: MIT

Welcome to Desearch powered by Bittensor Subnet 22! Desearch is a decentralized, AI-powered search engine that returns unbiased and verifiable results across X, Reddit, Arxiv, Hacker News, Wikipedia, YouTube, and the broader web. Frontend and API access are available at desearch.ai.

Table of Contents

Introduction

Desearch delivers an unbiased, verifiable search experience built on the Bittensor network. Miners compete to return the best search results from multiple sources; validators independently verify result quality and assign on-chain rewards. Through the public API, developers and AI builders integrate real-time, decentralized search into their products.

Key Features

  • AI-powered analysis — decentralized models produce relevant, contextual, unfiltered results.
  • Diverse data sources — X, Reddit, Arxiv, Hacker News, Wikipedia, YouTube, and general web.
  • Sentiment and metadata analysis — captures emotional tone and key metadata for social content.
  • Verifiable rewards — validators independently scrape and score miner outputs.
  • Extensible — community-driven improvements to scoring, sources, and relevance.

High-Level Architecture

  • Miners run a single Bittensor axon that answers IsAlive and all search synapses (AI / Twitter / Web). Validators call the axon directly via dendrite.
  • Validators generate synthetic queries every UTC hour, dispatch them to miners, independently verify results against ground-truth scrapers (Apify, ScrapingDog), and write weights on-chain. They also expose an organic-search FastAPI that the Desearch product backend calls on behalf of API consumers.
  • Bittensor network — settles miner compensation on-chain in $TAO.

Getting Started

For API Consumers

To integrate Desearch into your product, visit desearch.ai and request an API key. Consumers send requests to the Desearch API with their API key; the Desearch backend routes those requests to validators on your behalf and returns the aggregated search results.

For Miners

Miners contribute search capacity and are rewarded based on result quality and volume. Expected setup steps:

  • Prepare a server with Python ≥ 3.10, PM2, and a registered hotkey on netuid 22.
  • Configure credentials for OpenAI, SerpAPI, and Apify.
  • Declare per-search-type concurrency in neurons/miners/manifest.json.
  • Run the axon with PM2.

See the Miner Setup Guide for full instructions.

For Validators

Validators verify miner outputs and write weights on-chain. Expected setup steps:

  • Prepare a server with Python ≥ 3.10, PM2, Redis, jq, and a registered validator hotkey.
  • Configure credentials for OpenAI, Apify, ScrapingDog, and W&B.
  • Generate a public API access key and run the autoupdate script.

See the Validator Setup Guide for full instructions.

Additional Guides

Monitoring

Validators stream metrics to Weights & Biases. Public dashboards are available at wandb.ai/smart-scrape/smart-scrape-1.0.

Contact and Support

About

Subnet 22

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors