The cloud should not own the means of cognition. Ancapex Research turns decentralized compute into public model formation: training runs, evals, provenance, and ownership design for models communities can inspect, fork, and govern.
Launching with Model Run #1. No artifacts, no claim.
If you cannot inspect it, fork it, or run it, you rent it.
Intelligence is becoming infrastructure.
Infrastructure becomes dependency.
Dependency becomes governance.
We reject cognitive landlordism.
AI for all is not access.
AI for all is ownership.
No gods in the cloud.
AI is moving from tool to infrastructure. Infrastructure becomes dependency. Dependency becomes governance.
The issue is not that companies build models. The issue is enclosure: models people cannot inspect, fork, run, evaluate, or govern.
Ancapex focuses on model formation, not rented inference. The enemy is not a logo. The enemy is dependency without exit.
If intelligence is only available through closed APIs, access can be priced, filtered, throttled, or removed.
Inference gives access. Training shapes capability, values, memory, and institutional control.
A public model run needs logs, evals, provenance, and failure records. No artifacts, no claim.
Model Run #1 is the founding act of Ancapex: a public attempt to turn decentralized compute into a community-governed model artifact.
Status: forming run cohort.
Compute providers, researchers, evaluators, dataset stewards, protocol communities, funders, and narrative allies.
Compute qualification -> run design memo -> training / post-training recipe -> public run ledger -> eval report -> model card -> ownership memo.
A model artifact, public records, evals, failure notes, and a reusable template for future community-owned model runs.
Not a frontier-model claim. Not a token sale. Not an investment scheme. Not decentralization theater.
The cognitive proletariat is not a demographic. It is everyone whose work, memory, creativity, and judgment increasingly depend on models they cannot inspect, fork, run, or govern.
A movement is not a slogan. It is resource allocation: compute, researchers, datasets, capital, governance, and attention moving into shared model formation.
Design recipes, evals, refusal matrices, model cards, and failure reports.
Turn idle or fragmented capacity into qualified training substrate.
Sponsor, govern, and use models that reflect a real constituency.
Fund public model runs, infrastructure, provenance, and ownership design.
No permission from gatekeepers. Real markets for compute. Capital expenditure into public model formation.
Not anti-capital. Anti-monopoly. Capital against cognitive enclosure.
The homepage roadmap is deliberately narrow: ship the governance, baseline, public run, and training substrate required for a credible first run.
Define ownership, governance, licensing posture, and no-token-sale boundaries before public claims.
Start from open weights and make behavior legible before running public post-training.
Run a public post-training experiment with recipes, checkpoints, a run ledger, and failure records.
Turn decentralized GPUs into qualified training substrate with scheduling and verification rails.
We do not ask for trust before records. These are the first artifacts Ancapex is preparing for Model Run #1.
Access is not ownership. A model people cannot inspect, fork, run, or govern is rented cognition.
Training claims require logs, recipes, evals, checkpoints, and failure notes.
Datasets, licenses, provenance, and exclusions must be documented before launch.
Ownership, governance, fork rights, and usage rights must be defined before public claims.
Proof comes first.
We are launching. Bring a real piece of the machine.
Target: 10-20 serious providers for qualification. Bring GPUs, clusters, scheduling capacity, or infrastructure experience.
Target: 5-10 researchers for recipes, evals, model cards, refusal matrices, and failure reports.
Target: 3-5 communities willing to sponsor, govern, or co-own a public model run.
Bring licensed, public, or community-relevant datasets with clear provenance.
Fund public runs, compute qualification, evaluation infrastructure, and model commons design.
Help explain why ownership, not access, is the political question of AI.
Community-owned does not mean costless. Public model runs can be funded by sponsors, protocol treasuries, research grants, compute partners, foundations, and sovereign AI pilots.
Ancapex may receive operator fees for coordination, training infrastructure, evaluation, provenance, and governance design. No token sale. No implied investment return.
We respond directly where there is a credible fit for Model Run #1, compute qualification, research, funding, or governance work.
REF · —