Tribela Labs is live — building in the open

Tribela Labs is now public. The website, research publications, governance framework, and product development details are accessible from today.

The initial release covers four areas:

Research. The first paper — Efficient Content Moderation via Contextual Curriculum Learning — presents findings on fine-tuned transformer models for content moderation. The results indicate that these models can match or exceed LLM-based approaches while consuming substantially less energy. The full paper, data, and methodology are published openly.

Governance. A draft community governance model has been published for public review. The model proposes a one-person-one-vote structure with five stakeholder groups, contribution-based voting rights, and clear separation between operational and constitutional decisions. It is a first draft and is open for community feedback before formal adoption.

Products. Two products are in development. Guardrail is a cloud-based content moderation API. It uses classifier models to detect harmful content — harassment, abuse, and targeted harm — and returns transparent score breakdowns with open evaluation sets. The moderation policy is designed by Tribela Labs and published in full; model cards, evaluation suites, and policy changes are documented openly. Wyvern Engine is an on-device AI runtime scheduled for Q3 2026, designed with a cryptographic trust chain that records what ran, where, and what it observed. It operates in a memory-isolated environment with no external data transmission, meaning data does not leave the device. The target specification is 2B parameter models running in 2-3GB RAM.

Open problems. Six open research questions have been published across cryptography, fairness, infrastructure, and governance. These include inference integrity verification without weight access, cross-lingual harm taxonomy alignment, and verifiable federated training protocols. Contributions and discussion are welcome.

The research is published openly. The governance model is a draft open for review. Development of both products is ongoing. Contributors with experience in multilingual NLP, on-device inference, ZK cryptography, and AI policy are invited to participate.

— Tribela Labs · EU · EST. 2026