Founder & problem
Why an independent founder is building Meo
Ngoc is building Meo Meo AI as a personal, self-funded project in Vietnam. The motivation is practical: an affordable assistant should help with real work across personal devices, preserve useful context and research, and ask for permission before consequential actions.
The initial audience is individuals who want continuity across research, software work and everyday workflows without giving a model unrestricted control. Development starts with the founder’s own PC and Android workflows; this is not a claim of a public customer base.
The project is bootstrapped and actively developed. There is no registered company yet, no investors and no institutional funding. Affordability is a design goal, not a published pricing promise.
Product & architecture
One Meo, several environments
Meo is a personal AI operating layer, not a replacement operating system. Its direction is one coherent assistant with replaceable intelligence providers, shared rules and device-specific capabilities.
- PC Prime Core — foundation / internal proof of concept. A local runtime and transactional-data experiment has been built and tested. The PC is the intended long-term home for canonical coordination; this does not mean full PC production authority or general-purpose autonomy is complete.
- Android companion — implemented internal builds. Android has local data, conversation and review foundations, with recorded device acceptance for scoped checkpoints. In the long-term PC ecosystem it acts as a companion, while retaining useful standalone local behavior.
- Device / Meo Remote — experimental foundation and roadmap. Controlled interaction should use explicit human approval and bounded capabilities. Unrestricted remote administration is not available from this website.
- Research Room / PNC — bounded internal prototype. Source collection, research history and multi-model critique support comparison and verification. A live multi-model research proof of concept has recorded technical acceptance; that is not a guarantee that its conclusions are true.
- Modular practical applications — research / roadmap. Focused applications such as Meo Tiệm are being explored. There is no public application catalog or production maturity claim.
- Local-first / offline survivability — implemented local foundations, broader direction. Local data access must not depend on a cloud model. Cloud-dependent reasoning can be unavailable offline; Meo must report limits honestly rather than invent an answer.
Human approval stays outside the model. In the Teacher → Core → Tool approach, an AI Teacher offers advice or candidates, the local Core validates them, and tools act only within authorized scope. Cloud infrastructure supports the product; it does not become its authoritative brain.
Claude: historical role & intended use
Independent review helped shape the foundation
The Shared Core v0.3 consolidation record, dated 16 August 2026, explicitly incorporates Claude v0.2 architectural feedback. Its documented changes clarify shared-core boundaries, data integrity and governance. This supports describing Claude as an early independent AI architecture reviewer, alongside other AI-assisted work and the founder’s decisions.
The Multi-AI Teacher Council policy treats Claude and other models as advisers, not authorities. The original policy PR was later closed as superseded; it was documentation, not a runtime integration. Source-precedence corrections and the V1 architecture review record also preserve a distinction between design, verified code and Owner approval. Those records do not establish that every correction or V1 decision came from Claude.
Claude is not a cofounder, investor, official Anthropic partner or sole creator of Meo’s foundation. Historical review does not establish a production Claude API integration; none is claimed here.
Why Claude matters next
- Code and architecture review, including cross-device consistency.
- Independent security critique and failure-mode analysis.
- Research critique, source comparison and structured reasoning.
- Structured assistance within Teacher → Core → Tool boundaries.
Claude Startups support would help a self-funded founder evaluate models, test a bounded provider adapter and compare quality, failure behavior and costs. Any future integration remains subject to Owner authorization, validation, applicable API terms and an affordable operating budget. No support, eligibility or acceptance is implied. Meo’s local authority and provider replaceability remain unchanged.
Recorded milestones
Progress, with limits attached
Reviewed on 9 October 2026. Product records below are internal and the core repository is private. These are summaries of recorded commits, PRs, tests and build metadata—not a public independent audit. Private source, research conversations and operational details are not published.
- August 2026 — architecture and governance. Shared Core v0.3 incorporated Claude review feedback. Project-source corrections clarified which sources govern decisions and why design documents cannot prove implementation.
- 2 September 2026 — unified V1 architecture. An approved architecture record aligned one assistant experience with existing local data, provider and human-review boundaries. This milestone was a design decision, not delivery of every planned capability.
- 10–13 September 2026 — shared foundations and PC proof of concept. Shared-contract and transactional-runtime work was merged. Internal evidence records Windows/Android parity checks using disposable data; it does not establish production-scale reliability or power-loss safety.
- 30 September 2026 — scoped Android acceptance. An exact-source Android build, SHA-256 delivery metadata and runtime acceptance records support a data-preservation checkpoint. The artifact remains internal, not a public download.
- 2 October 2026 — live multi-model research experiment. Bounded Research Room technical acceptance and focused CI were recorded, followed by integration PRs. Research quality remains uncertain; this does not qualify a production credential foundation or activate PC authority.
- 8–9 October 2026 — public cloud foundation. A separate static website deployment provides HTTPS, status, documentation and release metadata. Public website baseline and deployment history are available for inspection.
There are no public binary releases yet. Project status describes development areas, not uptime promises or a certification of the whole platform.
Next milestones · Planned
From bounded foundations to dependable use
- Strengthen local PC continuity, recovery and data-preservation evidence.
- Validate narrow PC–Android workflows and permissioned device interaction.
- Evaluate a Claude adapter with scoped tasks, privacy-aware inputs, cost limits and failure tests before any production use.
- Prepare a safely distributable build with clear release notes and verification metadata when ready.
These are direction-level milestones, not promised dates or authorization to activate sensitive features. Owner Portal, unrestricted administration, private cloud memory and production signing remain off.