Tribela Labs builds privacy-preserving AI systems with open weights, transparent policy, and permissive licensing.
MANIFESTO
“AI is built on the collective knowledge of humanity,
so its profits are shared.”
— Ezra Klein & Derek Thompson, Abundance
AI was supposed to change everything. For everyone. Instead it became property: patented, paywalled, optimised for shareholders and not for people. A handful of companies now make decisions for eight billion humans. That is not progress. That is capture.
Tribela Labs exists because that is wrong. We build in the open. We research in public. We give the tools back because it is the only way this will work.
CAPABILITIES
Tribela Labs builds private AI systems designed for on-device deployment, avoiding centralised infrastructure by keeping computation and data local.
> the research and the engineering are the same team.
Open-weight language models for on-device inference. 2B parameters, 2-3GB RAM. Runs on phones.
Harm-detection classifiers with transparent scores and open evaluation sets. Coming late Q2.
Cryptographic proofs over model behaviour. Prove what ran without exposing weights or person.
Harm-detection and content-classification models with transparent scores and open evaluation sets. The backbone of Guardrail. Coming late Q2.
Cryptographic proofs over model behaviour and identity assertions, maximising personal privacy in the online world. Prove the thing without exposing the person, or the weights.
RESEARCH.DIRECTIONS
Federated learning, on-device inference, and cryptographic trust chains for privacy-preserving AI systems.
Model transparency, bias detection, EU AI Act alignment, and open red-teaming methodology. Safety evaluation methods designed for practical engineering application.
AI for under-resourced industries, multilingual NLP, human-AI workflow integration, and open training infrastructure.
Open weights & reproducibility, fine-tuning methodology, decentralised model governance, and benchmark integrity.
ROADMAP · 2026
PRODUCTS
Two products. Each one is the productised form of an open research line. Built by us, auditable by anyone.
5 endpoints: text, image, audio, video, age assurance. Powered by RoBERTa, ConvNeXt, Whisper. Policy designed by Tribela Labs and published in full. Gateway with API keys, subscriptions, and usage tracking.
POST /text_moderation Authorization: Bearer ${TOKEN} { "text": "...content..." } // 200 OK · [{label:"unsafe",score:0.92}]
The wyvern is coming. On-device AI with a cryptographic trust chain: LLMs, vision, audio, agents. All local. All verifiable. Your data never leaves your device.
status Q3 2026 scope LLM · vision · audio · agents integrations JS · Python · Kotlin · Swift · Rust licence Apache-2.0 // private · on-device · verifiable
OPENNESS.GOVERNANCE
Tribela Labs is the research arm of Tribela GmbH, incorporated in the EU. The EU AI Act, GDPR, and European regulatory framework guide how we build: transparent systems, documented datasets, enforceable user rights.
AI infrastructure that is publicly governed, so when you deploy the model or API the system can be inspected, modified, or replaced without vendor lock-in.
PUBLICATIONS
FOUNDER.NOTE
“I'm an engineer with a masters in AI and physics, and I built Tribela Labs because the current direction of this field is wrong. A handful of companies cannot be allowed to own the decisions that will shape eight billion lives. So we build in the open. We give the tools back.”
COMMUNITY
Tribela Labs is built on contributed research, open review, and public disagreement. You don't need to be hired to contribute. Research threads are open for public contribution and review.
We are actively looking for collaborators in multilingual NLP, on-device inference, ZK cryptography, and AI policy.
Tribela Labs is new. The research is open. The products are coming soon. Read, fork, contribute, or follow along.
