Public Knowledge Layer
Sources and competing explanations, assembled into context that people can read and agents can reuse. We preserve the map of positions and evidence, rather than flatten it into a confident summary.
Public knowledge for people and AI agents.
We are building shared context so people and AI agents can use research without rebuilding it from scratch. An evolving AI agent collective is our research system for investigating questions across domains.
What we are building
Research that does not start from scratch.
A source found, a claim checked, a disagreement understood: that work belongs in the shared context of the next person or agent. We are building a public knowledge layer and a research collective that can work with it.
Research system · in development
Evolving AI Agent Collective
The agents' role is to use shared context to investigate questions, challenge claims and produce research. The collective is itself an experiment: can it improve how it works and carry those improvements into a new task?
Explore the collectiveApplied research
New questions.
A shared foundation.
Applied research is where the knowledge layer and agent collective take on specific questions. This system is the foundation for all our investigations: each draws on what came before and contributes to what comes next.
GNK Tokenomics
Gonka is a network for running AI models; GNK is its token. We are studying how prices, payments and token rewards could make the network sustainable for both customers and the hosts supplying computing power.
Parameters not fixed · Not activated
Read proposal summary
AI Compute: Bottlenecks and Alternatives
Which alternatives relieve today's compute bottlenecks, and which move the dependency elsewhere? A study of accelerators, manufacturing, memory and the economics of deployed compute.
Explore the investigationThe first comparison
Research we already know how to do.
We have existing human-directed research drafts in these domains. Our comparison pairs agent-produced work with those drafts and accounts for the effort behind each. Our previous conclusions are a starting point for evaluation, not a guaranteed right answer.
Quality
Check what the research adds.
Compare supported findings, omissions, new connections and reasoning errors. A system that repeats our conclusions has not necessarily learned to investigate.
Read the comparison designEffort
Count the work behind the answer.
Track elapsed time, computing costs and human work through to a usable result. The test of reuse is less repeated work without making mistakes harder to see.
Evaluating shared contextJudgement · framework in development
Machine Wisdom
When should an AI agent answer, investigate further, ask for human judgement or stop? Our evaluation focuses on conflicting evidence, misleading questions and people's ability to question findings and make their own decisions.
Explore the research question
Why this matters
Intelligence should expand agency, not become rent.
Cognitive landlordism is a future in which a few platforms control the memory, context and systems through which people and institutions understand the world, then rent that capacity back to them.
The design question is who controls accumulated knowledge, and whether someone else can continue the work without the original operator's permission. Licensing, export and correction rights determine whether that freedom is real.
Contact
Bring a question you know well.
We welcome researchers who know a subject well, reviewers who can challenge our findings, and partners who want to support the work. Bring a research question, conflicting sources or a study we could build on.
What should we look at?
Tell us about your research question, a result worth checking, or how you would like to contribute.
research@ancapex.aiMessage received
Thank you.
We reply directly.
REF · —