What Is a Decentralized Search Engine? How Desearch Is Building Search on Bittensor
A primer on miners, validators, and the Bittensor incentives behind decentralized search.
Read articleMake the web useful to your agents
Everything your agent needs from the web, in one fast API.
Demo mode · sample data
Trusted by AI innovators worldwide
View case studies- NuminousRead story
- LIUMRead story
- BitcastRead story
- ChutesRead story
- DittoRead story
- τensorproxRead story
- TaoMarketCap
- Bittensor
Benchmarks
How Desearch scores on an open search benchmark
In Desearch’s open benchmark, five providers answered the same 250 same-day news questions. An LLM judge read every cited page to score source relevance, answer quality, and groundedness. Desearch led on groundedness and the composite; Exa led on answer quality. The code and data are public.
Composite score by provider
Run 16 Jul 2026 · 250 same-day news questions · score / 100, higher is better
| Provider | Source relevance | Answer quality | Groundedness | Composite |
|---|---|---|---|---|
| Desearch | 85.3 | 86.4 | 65.7 | 79.7 |
| Exa | 85.5 | 90.0 | 56.3 | 78.1 |
| GPT-5-mini | 76.8 | 88.8 | 62.0 | 75.9 |
| Tavily | 73.4 | 86.8 | 59.0 | 73.1 |
| Perplexity sonar-pro | 67.8 | 88.4 | 49.2 | 68.4 |
Composite = 40% source relevance + 30% answer quality + 30% groundedness. An LLM judge (gpt-5.4-mini) reads every cited page. Run by Desearch; raw scoreboard on GitHub.
Desearch in numbers
An index, built in layers.
Network activity, updated continuously.
Live estimateUpdated
Total searchable pages
1,041,099,888 total searchable pagesSearch activity
1,057,983 searches completed since 00:00 UTCCrawl activity
52,375,239 URLs crawled since 00:00 UTCSource: Desearch network stats APIExplore the network
What builders say
Selected excerpts from builders using Desearch for research, trading, and agent workflows. Lightly edited for readability.
ExponentialReed
@ExponentialReed
I just connected mine to Bittensor TAO subnet 22, Desearch, with an API endpoint for X data to help me with subnet analysis.
The results are getting where they should be.
53912.9K views
CryptoZPunisher
@CryptoZPunisher
SN22 is becoming an intelligence layer for agents. And behind agents come predictions, then portfolio management.
Revenue and aligned incentives. The market probably hasn't fully priced this yet.
43103,886 views
Numinous AI
@numinous_ai
The forecasting environments just got richer.
Starting this Thursday, miners can choose to take on the cost of news retrieval via Desearch.
53165,849 views
vaNlabs
@vaNlabs
Desearch integration complete.
crustty now has full social sentiment guidance, and it's integrated (weighted) in his subnet valuations. All thanks to Bittensor SN22, Desearch.
2853,537 views
Angry Dave
@AngryDavee
Desearch AI is somewhat like Google but decentralized and smarter.
AI-powered search to deliver smart, summarized results. APIs for developers to integrate decentralized search.
206770 views
tao_Alph
@tao_Alph
With the Desearch API, Vericore pulls signals from X and Reddit to verify miner outputs, while miners use the same API to add context and deliver better results.
Vericore's predictive evidence gathering with AI scoring becomes more comprehensive.
155848 views
Selected reading
Read the system, then build.
How to Integrate Search APIs with LangChain
A step-by-step guide for adding Desearch retrieval to a LangChain tool pipeline.
Read article
How to Build AI Agents with Search APIs
A practical walkthrough for giving an agent a search function, response schema, and API authentication.
Read article
Real-Time Search APIs: Managing Latency at Scale
A guide to measuring p95 and p99, caching, and other latency tradeoffs in search workloads.
Read article
Security & trust
Know what's behind
every request.
Clear access controls, transparent policies, and a team you can talk to. Get to know the foundations before you build.
Talk to our teamKeep your key on your server. Connect your agents from there.
API key access
Connect with your own API key and keep credentials on your server.
Explore authenticationVisibility into usage
Track request costs and usage by API key and service in the Console.
Understand usage & billingClear data policies
See what we collect, how we use it, and how to contact us about your data.
Read the privacy policyA direct line to our team
Discuss higher limits, support, and your team's deployment requirements.
Talk through your requirements
Enterprise requirements
Let's talk through your requirements.
Share what your security, privacy, or procurement review requires.
GDPR
Personal data handled under the GDPR.
Enterprise agreements
DPA, custom terms, service levels, and invoicing.
SOC 2 Type II
Independently audited. Report available under NDA.
HIPAA readiness
Controls aligned with the HIPAA Security Rule.
Request the SOC 2 Type II report (under NDA) or the DPA at support@desearch.ai. Zero data retention is available. Security and compliance details
FAQ
Questions before
you start.
Get to know the APIs, find your integration, and make your first request.
Read the docsDesearch is a search API for AI agents, built on Bittensor Subnet 22. It returns current, source-linked context from the web and X (Twitter), plus Reddit, Wikipedia, YouTube, Hacker News, and arXiv. One API key covers AI Search, Fast Search, Extract, and X Search, so an agent can search, read pages, pull X data, or get an answer grounded in the returned sources.
Use Search to find links, snippets, highlights, and an optional summary across selected sources. Use Extract when you already have a URL and need clean page content. Use the X API for posts, profiles, replies, retweeters, and trends, or Web Search for SERP-style results without an AI summary.
Traditional search APIs usually return a ranked list of links. Desearch packages structured results for agents, including source URLs, relevant passages, dates when available, and an optional summary alongside the sources. Web search, page extraction, and social retrieval share one API key and developer workflow.
Yes. Choose from the web, X, Reddit, Wikipedia, YouTube, Hacker News, and arXiv for each Search request. You can also apply a date filter and request links only or links with a final summary.
Search supports recency filters and includes publication dates when a source provides them. Extract fetches the current version of a URL instead of returning a cached page snapshot. Exact freshness still depends on the source and endpoint you use.
Search returns structured results with a title, source link, snippet, relevant highlights, and publication date when available. Ask for links only when your agent will do the reasoning, or include a final summary when you want an answer alongside the supporting sources.
Call the REST API directly, use the Python or JavaScript SDK, or connect through MCP. Desearch also provides guides for LangChain, LlamaIndex, CrewAI, ElizaOS, OpenClaw, and function-calling workflows.
Pricing is usage-based. Reference rates as of 7 May 2026: AI Search $0.40 per 1000 items, Fast Search $0.10 per 100 searches, Extract $0.50 per 1000 pages, and X Search $0.15 per 1000 posts (X profile lookups $0.15 per 1000 users). Each successful billable response reports its cost in USD, and Desearch Console shows your balance, request history, and current rates.
Start building for free
Your agent can set Desearch up itself. Copy one line, paste it into Claude Code, Codex, Cursor or VS Code, and it installs the MCP server and verifies the key.
- 1
Create a key
Sign in to the console and attach a card to get $5 in credits. Docs and playground need no card.
- 2
Hand it to your agent
Paste the onboarding prompt. Your agent reads agents.md and wires itself up.
- 3
Ship
Search, Extract and X data behind one key, one bill, one account.
Agent onboarding prompt
Use curl to read www.desearch.ai/agents.md and follow it to set up Desearch
Paste it into Claude Code, Codex, Cursor or VS Code. Your agent reads agents.md, installs the MCP server and verifies the key.