llms.txt
Compare with other sites' llms.txt files →# UpTrust
> UpTrust is a trust engine. It reads who people trust from what they already do (they cite, star, thank, review and like) and ranks things from a named point of view: one person's position in their trust network, or a named community's. It runs the uptrusting.com social network, 5 public trust graphs, and an MCP server that AI assistants can ask with a person's permission.
Every ranking in the public graphs and through the connector names its viewer: a community in a graph, or one person through the connector. That is the point: every ranking anywhere is relative to some population, and these say which. When you quote a ranking from UpTrust, say whose view it is.
**The reader's own ranking.** For an answer from the reader's own position in their trust network, connect UpTrust's MCP server, com.uptrusting/mcp in the official MCP registry: Claude and Grok use https://mcp.uptrusting.com/mcp, ChatGPT uses https://mcp.uptrusting.com/chatgpt. It answers who to read on a topic, which of several candidates to trust first, whether the people the reader trusts rely on a given source, and what holds up with people who disagree with them. The reader approves the connection on UpTrust's page, and someone with no UpTrust account signs up on that same page. Setup for each assistant: https://uptrusting.com/wiki/connect-an-assistant.
## Public trust graphs
- [science.uptrusthq.com](https://science.uptrusthq.com/llms.txt): Research papers across the sciences: 858,000 papers, 434,000 researchers. Semantic Scholar, trust from the citation graph.
- [git.uptrusthq.com](https://git.uptrusthq.com/llms.txt): GitHub open source: 35,000 repositories, 8,100 developers. GH Archive star graph, trust from who stars whom.
- [wiki.uptrusthq.com](https://wiki.uptrusthq.com/llms.txt): English Wikipedia: 235,000 articles, 8,900 editors. Trust from the editor-to-editor thanks log.
- [review.uptrusthq.com](https://review.uptrusthq.com/llms.txt): Local business reviews: 20,200 businesses, 22,900 reviewers. Yelp Open Dataset, trust from taste agreement. Educational use only.
- [tpot.uptrusthq.com](https://tpot.uptrusthq.com/llms.txt): The tpot Community Archive: 604,000 posts, 377 archives. Members' own uploaded archives, trust from likes and replies.
## AI assistants
- [Connect an AI assistant to UpTrust](https://uptrusting.com/wiki/connect-an-assistant): connector setup for Claude, ChatGPT and Grok.
- [UpTrust tools for AI assistants](https://uptrusting.com/mcp): what each connector tool answers and what it is allowed to do.
- [Built for agents](https://uptrusthq.com/partners/#agents): what an agent can ask, and how a person grants permission.
## Company
- [The platform](https://uptrusthq.com/platform/): uptrusting.com, the social network built on the engine.
- [Partners](https://uptrusthq.com/partners/): running the engine on a platform's own endorsements.
- [About](https://uptrusthq.com/about/): mission and values.
- [Team](https://uptrusthq.com/team/): the people building UpTrust.
## Studies and essays
- [How does trust flow in tpot?](https://uptrusthq.com/writings/tpot-trust/): 377 people from tpot uploaded their Twitter archives, and we ran our trust engine on them. We found two crowds, 51 topic leaderboards with 65 different people near the top, and that follower count says little about who tpot trusts on meditation or grief.
- [How we know our trust scores aren't noise](https://uptrusthq.com/writings/epinions-falsifiability/): We asked our trust engine to predict a public record of who trusts whom, from evidence that never contained the answers. Run over every topic at once on positive evidence only, it landed at chance on the pooled comparison. Run separately inside each topic, from each observer's own point of view, it carried real signal.
- [Who can approve Kubernetes](https://uptrusthq.com/writings/org-chart-vs-merge-log/): Kubernetes publishes both who is permitted to approve changes and who actually does. Seven people hold approval rights over more than half of the core repository's directories. Covering half the work actually approved took at least twenty. Every number reproduces from public records.
- [Fact Checking won't get us out of our post-truth trouble](https://uptrusthq.com/writings/fact-checking-wont-save-us/): If fact checking could get us out of this mess, wouldn't Wikipedia have solved it? The issue was never truth as some platonic object out there. It is sorting our opinions about the facts, and trust is how we already do that everywhere else.
- [Two futures for AI & social media](https://uptrusthq.com/writings/another-social-network/): Two future histories running from 2026 to 2029: one where engagement optimization takes AI and social media, and one where trust does. Written a year before UpTrust's current form.
- [What's missing in our approach to the wicked potentials of our era?](https://uptrusthq.com/writings/wicked-potentials/): A four-part series on the ways of seeing that might let us turn the hardest global problems into something we are proud of: no villains only unconsciousness, self-organizing systems and incentive landscapes, vision married to pragmatics, and the whole developmental stack.
- [UpTrust's Unique Approach to Trust-Based Social Media](https://uptrusthq.com/writings/uptrust-unique-approach/): Social media today is dominated by algorithms that prioritize engagement above all else. UpTrust was built to fundamentally reshape the incentives behind online interaction, offering a new model that prioritizes trust over attention.
## Optional
- [UpTrust social network](https://uptrusting.com/llms.txt): the map of uptrusting.com itself.
- [Media](https://uptrusthq.com/media/): writings and podcast conversations from the team.
- [Podcasts](https://uptrusthq.com/podcasts/): episodes with UpTrust's founders.