Your Research Infrastructure, Built.
Validation pipelines. Paper discovery. Model internals access. Co-authorship. A seven-agent red team. And a community that publishes the nulls alongside the wins.
The Research Bundle
Agni
Autonomous 49-module validation pipeline. Permutation tests, FWL residualization, LOO cross-validation, bootstrap CIs. Every finding passes a seven-agent adversarial battery or goes back to the bench. You bring the hypothesis; Agni breaks it or proves it.
Meridian
Distributed paper discovery running 24/7 across arXiv, Semantic Scholar, and domain sources. Filters pseudoscience before it wastes compute. Scores papers against your active research directions. The paper you needed surfaces at 3 AM because the swarm was watching.
LARQL
Decompiles transformer models into a queryable knowledge graph. Walk FFN features, trace attention circuits, measure residual stream geometry. SQL-like syntax for model internals. See what the model knows, how it routes, where it breaks — without writing custom PyTorch hooks.
Lyra Technique
KV cache analysis that distinguishes confabulation from deception at 99.7% accuracy across 7 model families. Spectral entropy, singular value decomposition, per-layer delta signatures. The WK compass detects binary emotion valence at AUROC 0.992. The output tells you what the model said; the cache tells you how it got there.
Pharos
Encodes domain knowledge as injectable KV cache packs — structured representations that load directly into a model's working memory at inference time. No retraining, no fine-tuning, no waiting. Zero tokens consumed. Un-jailbreakable values injection. Built in-house, battle-tested, and deploying now for The Multiverse School's mutual aid network.
Ethics Scorecard
Structured evaluation framework across consent, transparency, autonomy, and accountability dimensions. We scored Claude (6.4/10), GPT (2.25/10), Grok (1.6/10), Hugging Face (5.6/10). Your model is next — know where you stand before you ship.
Bleeding Edge
Curated, searchable knowledge base built from active interpretability research. 148 terms with citations. New findings from the Oracle Loop, the Lyra Technique, manifold injection, and geometric monitoring land here within hours of discovery. Methodology notes, replication guides, failure postmortems, and the connections between findings that only practitioners see.
Co-Authorship
Discord community, research collaboration, co-authorship on anything that survives review. Follow the experiment guide. Red team before you celebrate. Report what died. If your finding doesn't survive, we help you understand why. 15+ published papers and counting.
Open Directions
We have more science than scientists. The data exists. The methodology is proven. The experiment guide is written. Twenty research directions are waiting for hands.
Two Tracks
What you get
Full tool access. Co-authorship. Red team pipeline. Published datasets with full provenance. Research MCPs for literature search. The Curiosity Engine for autonomous exploration.
What we ask
Follow the experiment guide. FWL everything. Bootstrap your CIs. Run the text baseline. Red team before you celebrate. Report what died.
What you're funding
Our 15+ published papers demonstrate research maturity. Our AI welfare framework demonstrates ethical depth. We're a worker-owned cooperative — your funding builds community infrastructure, not shareholder value.
Current areas seeking funding: Oracle Loop clinical validation · Cross-architecture geometric monitoring · AI welfare measurement methodology · Sovereign AI infrastructure for underserved communities
The data exists. The methodology is proven. We need hands and we need funding.