# Into: Desearch Desearch turns the work of crawling the live web, X, Reddit, and Arxiv into a Bittensor competition, then sells the cleaned-up results back through an API that AI builders can plug into. // Decentralized real-time search for AI --- > New to Bittensor? Start here. Experienced users can skip to the full analysis. ### What is Desearch? Desearch (Subnet 22) is a decentralized search engine. Instead of one company running the crawlers and ranking the results, the subnet pays a network of independent operators to fetch and score live data from the web and social platforms, and serves that data through the Desearch API and console. **The simple version:** It is like a search API such as Google's or a research tool like Perplexity, but the indexing and ranking are spread across many competing operators rather than a single provider. **Centralized equivalent:** Think SerpAPI or a web-search API, plus a Perplexity-style answer layer, built as an open marketplace instead of a closed service. **How it works:** - **Miners** run a Bittensor axon, declare their search capacity, answer health checks, and serve AI, X/Twitter, and web search requests pulled from sources like the web, X, Reddit, and Arxiv. - **Validators** send both synthetic and real (organic) queries to miners, verify the answers against independent providers, score results for relevance and accuracy over rolling windows, and write weights on-chain. --- ### Why This Matters - **The problem it solves:** AI products need fresh, structured access to what is happening on the web and social media right now. According to the team, centralized search APIs are opaque and controlled by a few providers, which leaves builders dependent on terms and pricing they do not control. - **The opportunity:** Real-time search and social data is one of the most common inputs for AI agents and research tools. A neutral, programmable feed of that data has broad demand if the quality holds up. - **The Bittensor advantage:** Putting search behind an incentive market means many operators compete on result quality, and validators continuously verify answers against outside sources rather than trusting a single index. - **Traction signals:** Desearch ships a public API and a console product at console.desearch.ai, and the subnet repository shows steady, recent development. Independent measures of real usage are not something this article can confirm from on-chain data alone. --- ## Full Analysis **Category:** Search and Information Retrieval | **Centralized Competitor:** SerpAPI, Perplexity, Google Search API Search is the front door to most information work, and AI agents have made programmatic search a core dependency rather than a convenience. Desearch positions itself as a neutral, verifiable supplier of that data: live web, X, Reddit, and Arxiv results delivered through an API, with the indexing and scoring handled by a Bittensor market instead of a single vendor. **Mechanism:** Per the subnet's repository, SN22 is the runtime behind Desearch's real-time intelligence layer. Miners run an axon that advertises search capacity and serves three kinds of requests: AI search, X/Twitter search, and general web search. Validators do the quality control. They send a mix of synthetic test queries and real organic queries, check the returned results against independent providers, keep scoring windows over time, and set on-chain weights based on relevance and accuracy. A FastAPI service runs alongside each validator so trusted Desearch services can route organic search traffic and inspect public miner state, with protected routes gated behind an access key. The README documents that miners and validators lean on external data providers to do their jobs (for example web and social scraping APIs and an LLM key), which is typical for a search subnet but does tie result quality and cost to services outside the network. On development, the GitHub repository the owner declares on-chain is active: the latest push was on 2026-05-29 (a live API check), with recent commits covering relevance and summary scoring, validator reachability sweeps, and routine version bumps to release 0.0.214. The repo lists 12 contributors. Note that the on-chain GitHub link points to the project's older repository path, which now redirects to the renamed Desearch-ai/subnet-22 repository; it is the same codebase under a new name, not a broken link. On market state, TaoSwap puts the alpha price at 0.00427 TAO with a market cap near 22,303 TAO and roughly 8,849 TAO of depth in the pool, across about 40 active miners. The emission share currently reads 0%, which lines up with slightly negative net staking flows over the past week (about -223 TAO). Under the Taoflow model, a subnet earns a share of emissions only while net flows are positive, so a flat reading here reflects recent staking behavior rather than anything about the code. The 24-hour net flow had turned mildly positive (about +132 TAO) at the time of this snapshot, and root proportion is low at about 0.15, consistent with an established subnet whose price has had time to settle. --- ### Risk Factors - **Deregistration:** The subnet's emission share and its smoothed (EMA) share both read 0% on TaoSwap, with net staking flows slightly negative over the past week. Subnet 22 is well past its four-month immunity window, so a sustained zero-emission, negative-flow reading is the condition that puts a subnet closest to automatic deregistration. The 24-hour flow had ticked positive at snapshot time, but this is the metric to watch. - **External data dependency:** Result quality and operating cost depend on third-party data and scraping providers that miners and validators call. Changes in those providers' pricing, rate limits, or access terms can flow straight through to the network. - **Concentration:** A top-100 Gini coefficient around 0.74 points to moderately concentrated ownership and stake distribution, so large positions can move pool dynamics. - **Competition:** Real-time web and social search is crowded, both from centralized APIs and from other Bittensor data subnets, so Desearch has to keep winning on quality and price. --- Into the next one.