Fenrua-V1

Fenrua V1

Public development preview · closed weights · controlled hosted access planned · no inference currently released

Publication boundary
Fenrua V1 is the public name of Fenrua Labs’ first general-purpose model family. This repository is a development record only. It does not currently contain downloadable weights, a frozen architecture, a completed training run, benchmark evidence, a public API, a mobile application, or a production service.

Naming

public_model_family: "Fenrua V1"
developer: "Fenrua Labs Pty Ltd"
internal_programme_codename: "FML-Mosaic"
legacy_public_identifier: "FML-Mosaic-527B"

FML-Mosaic remains a development and engineering codename. The public product family is:

• Fenrua V1 Mini
• Fenrua V1
• Fenrua V1 Pro

The earlier FML-Mosaic-527B name was a development identifier and target-scale reference. It must not be interpreted as proof that a trained 527-billion-parameter checkpoint currently exists.

Current status

Mission

Fenrua V1 is being developed as a broad, practical, evidence-aware assistant family for students, developers, researchers, builders, educators, professionals, and curious learners.

The intended assistant should:

• explain technology, coding, computer science, AI, mathematics, science, and engineering clearly;
• support academic learning, research, source evaluation, and digital literacy;
• distinguish fact, observation, inference, assumption, hypothesis, and opinion;
• challenge weak reasoning constructively rather than agree automatically;
• state uncertainty and missing evidence plainly;
• help users build practical modern skills; and
• use occasional clean humour when it improves learning without trivialising serious work.

Evidence before authority. Useful before flashy. Clear before complex.

Public product family

All names, access routes, and release dates remain development targets until candidate-bound evidence is published.

The community tier is intended to be genuinely useful. “Community access” means generous controlled use and contribution pathways—not a deliberately crippled demo and not a downloadable checkpoint.

Why Hugging Face

Hugging Face is Fenrua’s public community and research entrance for:

• model-family documentation;
• development status;
• evaluation updates;
• responsible-use boundaries;
• community contribution intake;
• future protected demonstrations; and
• public-safe research records.

It is not the intended public weight-distribution channel for Fenrua V1.

Planned access and Fenrua Compute Units

Fenrua intends to meter hosted model and platform services through Fenrua Compute Units (FCU).

FCU_public_boundary:
  service_accounting_unit: true
  active_now: false
  transferable: false
  cash_redeemable: false
  tradable: false
  wallet_asset: false
  investment_value: false
  appreciation_claim: false
  yield_or_staking: false

A future community launch programme may provide promotional FCU to eligible users and accepted contributors. Any public allocation figure remains a planning ceiling until serving benchmarks, anti-abuse controls, billing rules, and release approval are complete.

FCU will not represent raw text-token ownership. Future metering may account for model class, input and output volume, context length, modalities, tool use, memory, storage, and service priority.

Community contribution

Fenrua intends to reward verified useful contribution—not raw inference volume.

Potential eligible contributions include:

• high-quality, permissively licensed source discovery;
• accepted dataset cleaning and deduplication;
• provenance or licence corrections;
• reproducible evaluations;
• confirmed bug and safety findings;
• translations and accessibility improvements;
• educational material; and
• accepted tooling contributions.

Potential recognition may include promotional compute grants, bounded access increases, discounts, reputation, early-access eligibility, and research-grant consideration.

Automated self-conversations, duplicated workloads, token farming, or unreviewed uploads are not useful contribution evidence.

HuntingKnowledge

HuntingKnowledge is the knowledge-acquisition and preparation programme supporting Fenrua V1.

Priority areas

1. Programming and software engineering
2. Computer science and algorithms
3. AI and machine-learning literacy
4. Mathematics, probability, logic, and statistics
5. Physics, chemistry, biology, and earth science
6. Engineering, electronics, and robotics
7. Academic writing, study methods, and research skills
8. Source evaluation and evidence literacy
9. Digital literacy, privacy, and safe cyber awareness
10. Creative technology, communication, and practical productivity

Data and licensing boundary

Fenrua prioritises clearly reusable material, especially Apache-2.0 and MIT sources, plus separately reviewed educational or attribution licences where terms are clear.

Fenrua rejects or quarantines:

• unclear licence evidence;
• copyrighted material without reuse authority;
• personal information and secrets;
• private endpoints or infrastructure;
• duplicated, broken, spam, or SEO material;
• unsafe offensive-cyber material; and
• low-value content that does not improve the intended assistant.

Source-material licences do not automatically determine the licence or terms of a future dataset, model, API, or application. Each released artifact will receive its own explicit terms.

This model card grants no licence to Fenrua V1 weights, tokenizer assets, architecture files, training data, API access, or future applications.

Proprietary model boundary

Fenrua does not currently intend to publish:

• model weights;
• tokenizer assets;
• architecture configuration;
• exact corpus inventory or mixture ratios;
• cleaning and filtering recipes;
• training and optimisation methods;
• alignment data;
• system instructions;
• evaluation canaries;
• serving topology; or
• private model and infrastructure repositories.

Fenrua still intends to publish responsible-use information such as status, intended uses, limitations, candidate-bound evaluation results, high-level data categories, licence methodology, privacy boundaries, and version notices.

Encrypted vault direction

Fenrua V1 is planned to work with user- and organisation-controlled encrypted vaults.

Content off-chain. Commitments on-chain. Keys with the user or authorised organisation.

The planned direction is:

• prompts, outputs, projects, documents, and memory remain in encrypted off-chain storage;
• Chain 978 may carry bounded personal vault and continuity commitments;
• Chain N521 may carry bounded organisation vault, policy, and governance commitments;
• neither chain stores raw user content or acts as a public wallet; and
• no model, vault, or chain-linked service is made available by this card.

A conventional hosted model service normally must process the content a user deliberately submits in readable form during inference. Fenrua therefore does not currently claim cryptographically invisible hosted inference. A future service must publish its actual retention, logging, training-use, isolation, and confidential-compute boundaries before release.

Intended uses

Subject to future evaluation, access terms, and release approval, intended uses include:

• coding education and software-development assistance;
• STEM tutoring and conceptual explanation;
• AI and digital-literacy education;
• research planning and source evaluation;
• writing, organisation, and practical problem solving;
• developer experimentation through controlled APIs; and
• a general assistant that encourages stronger questions and reasoning.

Out-of-scope uses and prohibited claims

This repository does not establish that Fenrua V1:

• has completed training;
• contains a specific verified parameter count;
• exceeds any named model or benchmark;
• is safe or reliable for every audience or use case;
• is production ready;
• is available through an API or application;
• provides universal factual truth;
• proves semantic truth, legal authority, intent, or real-world effect;
• provides a public wallet, token, exchange, liquidity pool, staking, yield, custody, settlement, lending, remittance, or investment product;
• offers company shares; or
• operates as FML-521-A or as a production BlackBox Protocol service.

Evaluation and release evidence

Before any public release claim, Fenrua intends to publish evidence appropriate to the candidate, including:

• exact model and tokenizer version;
• intended use and limitations;
• benchmark methodology and raw results;
• coding, STEM, reasoning, and instruction-retention evaluations;
• safety, privacy, bias, and misuse evaluations;
• adversarial and regression testing;
• serving and capacity evidence;
• independent review status; and
• an explicit human-controlled release decision.

A successful local test does not become a public capability claim by itself.

Model extraction and abuse boundary

Future hosted access may include:

• authenticated accounts;
• per-user and per-model rate limits;
• bounded output lengths;
• no raw-logit, hidden-state, gradient, or weight access;
• extraction-pattern and repetitive-sampling detection;
• account and device abuse controls;
• model fingerprint testing; and
• terms prohibiting unauthorised extraction, reverse engineering, and distillation.

The public interface is intended to be useful, but it is not intended to expose Fenrua’s confidential model assets or training methods.

Related Fenrua programmes

These programmes share evidence discipline but are not interchangeable.

Official sources

• Website: https://fenrua.ai
• Hugging Face organisation: https://huggingface.co/Fenrua-Labs
• Company and research record: https://huggingface.co/Fenrua-Labs/FenruaLabs
• GitHub: https://github.com/Fenrua-Labs-Pty-Ltd
• Company X account: https://x.com/FenruaLabs
• Product X account: https://x.com/FenruaAI
• Contact: [email protected]
• Location: Sydney, Australia

---

Closed weights. Open interface. User-owned encrypted vaults. Verifiable continuity.