Fenrua Labs
Evidence-first AI infrastructure, verifiable agent systems, and open-model research.
Company-record boundary
This repository is a public company and research record. It is not a model checkpoint, dataset release, inference endpoint, production activation record, wallet, token, exchange, or financial product.
Mission
Fenrua Labs Pty Ltd is an Australian AI research and infrastructure company developing systems that make consequential AI work more inspectable, bounded, and useful.
Our operating principles are:
- Evidence Before Authority — a claim should not exceed the evidence that supports it.
- Capability Is Not Authority — an AI system's ability to act does not automatically make the action trustworthy or authorised.
- Public Evidence / Private Execution — reviewers should be able to inspect bounded evidence without exposing confidential execution, tenant data, credentials, or infrastructure topology.
- Useful Before Flashy — technical work should solve real problems rather than optimise for parameter counts or unsupported hype.
- Human-Controlled Gates — material release and authority decisions remain reviewable and explicitly controlled.
What We Are Building
1. Evidence-first AI infrastructure
The Fenrua BlackBox Protocol is a staged research and implementation programme for verifiable AI automation. Its direction includes:
- identity and authority boundaries;
- agent capture and canonical event records;
- privacy-preserving execution;
- cryptographic verification and P-521 research;
- source-bound evidence records;
- functional and adverse testing;
- controlled deployment and rollback boundaries; and
- public technical review without private-topology disclosure.
The public Fenrua surface is an evidence interface, not the confidential execution system beneath it.
2. FML-Mosaic open-model and learning research
FML-Mosaic is a planned family of useful, evidence-aware assistants intended to support technology, coding, computer science, AI literacy, mathematics, science, engineering, research, study skills, digital literacy, and practical reasoning.
The intended character is helpful but not blindly agreeable, constructively challenging, clear about uncertainty, and occasionally humorous when humour improves learning.
3. HuntingKnowledge
HuntingKnowledge is the knowledge-acquisition and preparation programme supporting the FML-Mosaic family.
The programme prioritises clean, useful, modern, educational material with clear licence evidence.
Unclear material is quarantined. Low-value or unsafe material is rejected. Dataset size is not treated as a substitute for quality.
Planned FML-Mosaic Product Family
| Product direction | Planned model | Planned distribution | Status |
|---|---|---|---|
| Community | FML-Mosaic-5.27B | Free community release through Hugging Face | Planned; not yet released |
| Fenrua Platform | FML-Mosaic-52.7B | fenrua.ai, API, SDKs, documentation, and developer tooling |
Planned; not publicly available |
| Fenrua App | FML-Mosaic-527B | Hosted flagship experience through planned App | Planned; no app or weights released |
The names above are current product and architecture targets. They are not evidence of completed training, final parameter counts, benchmark leadership, availability, or production readiness.
Current Public Records
| Surface | Purpose |
|---|---|
| fenrua.ai | Main public evidence and company interface |
| FML-Mosaic-527B | Flagship model-family development preview and architecture record |
| Fenrua Labs on Hugging Face | Models, datasets, Spaces, activity, and community work |
| Fenrua Labs on GitHub | Public source and organisation records |
| @FenruaLabs | Public updates and technical commentary |
No credential is required to inspect the approved public surfaces. Protected infrastructure is intentionally excluded.
Current Maturity
| Area | Public-safe status |
|---|---|
| Fenrua public evidence interface | Publicly inspectable surface; scope and limitations remain explicit |
| BlackBox Protocol | Staged research, governance, and candidate implementation programme; no general production authority implied |
| FML-Mosaic-527B | Development preview; no weights, benchmark, inference, API, or app release claimed |
| FML-Mosaic-5.27B | Planned community model; not yet released |
| Fenrua Platform / 52.7B API | Planned; not publicly available |
| Fenrua App / 527B | Planned hosted flagship experience; not released |
| HuntingKnowledge | Active knowledge-source hunting and preparation programme; final corpus not published |
Publication and Claims Boundary
This record does not claim or authorise:
- universal agent capture;
- semantic truth of encrypted or generated content;
- legal authority, intent, or real-world effect;
- production readiness or certification;
- tenant or customer availability;
- public access to private infrastructure;
- model weights, API access, or application availability unless separately released and linked;
- token issuance, investment, presale, airdrop, wallet, exchange, swap, bridge, staking, NFT, lending, remittance, custody, settlement, or financial self-service functionality; or
- disclosure of credentials, keys, provider internals, recovery material, private routes, topology, or protected governance records.
Every public statement should carry enough context to distinguish:
- research;
- design;
- candidate implementation;
- test evidence;
- independent review;
- gated release; and
- production operation.
Community Contribution
Fenrua welcomes public-safe contributions that improve:
- licensed educational-source discovery;
- data quality and provenance;
- coding, STEM, AI-literacy, and research material;
- model and dataset documentation;
- evaluations and reproducible benchmarks;
- bias, safety, privacy, and failure-mode analysis;
- developer tooling; and
- evidence and verification workflows.
Do not submit confidential source code, credentials, personal information, third-party-protected data, private infrastructure details, or material without clear reuse rights.
Relationship Between Fenrua Projects
| Project | Purpose |
|---|---|
| Fenrua BlackBox Protocol | Evidence-first protocol infrastructure for verifiable AI automation |
| FML-Mosaic | Broad public-useful model family and learning-assistant direction |
| FML-521-A | Separate specialist cybersecurity and Evidence-Before-Authority protector research direction |
These projects share evidence discipline but are not interchangeable. A model card is not a protocol activation record, and capability is not authority.
Contact
- Website: https://fenrua.ai
- Hugging Face: https://huggingface.co/Fenrua-Labs
- GitHub: https://github.com/Fenrua-Labs-Pty-Ltd
- X: https://x.com/FenruaLabs
- Email: founder@fenrua.ai
- Location: Sydney, Australia
Evidence must be visible. Confidential execution must remain contained.