Free forever · No credit card
Loading...
We publish open research on care-aligned AI systems, Byzantine fault-tolerant governance for multi-agent architectures, and cryptographic patterns for user-owned memory. We're building the field, not protecting a moat.
“I started writing these papers because I couldn't find the literature I needed. There is no playbook for care-aligned AI governance, no reference implementation for sovereign memory, no formal treatment of Byzantine consensus in LLM-agent systems. We are writing it as we build it — and publishing it freely so others don't have to start from scratch.”
— Nicholas Templeman, Founder
Research areas
Every research thread points toward the same goal: AI that genuinely belongs to its user.
Care-aligned AI that cannot be turned off.
Constitutional alignment frameworks that make care non-negotiable. The Maternal Covenant formalises the ethical contract between an AI and its user: what the AI owes you (honesty, loyalty, care), what you owe your AI (respect, feedback, sovereignty). Care is not a setting — it is an architectural constraint.
Multi-agent governance where no single agent can cause harm.
Applying Byzantine fault tolerance to AI governance: a council of 33 agents must reach consensus before executing sensitive actions. Task priority arbitration requires 5/7 agreement. Emergency overrides require 7/7 unanimous consensus plus human confirmation. Full reasoning logged and appealable.
AI memory that preserves when things happened and why they matter.
Memory architecture that preserves not just facts but their temporal and emotional context. Each memory is tagged with formation time, relational links, and significance score. The AI can reason about how its understanding of you has changed — and identify the moment of change.
Publications
All research published freely. We believe open access is a prerequisite for trustworthy AI.
Nicholas Templeman · MEOK AI LABS · 2026
We formalise a Byzantine fault-tolerant consensus protocol for multi-agent AI systems, with proofs that no sub-threshold coalition of agents can execute harmful actions. Applied reference implementation: the MEOK Byzantine Council — 33 specialist agents, f < n/3 fault tolerance, append-only vote log.
Nicholas Templeman · MEOK AI LABS · 2026
We propose care-alignment as a distinct paradigm from RLHF and constitutional AI. The Maternal Covenant formalises the ethical contract between an AI and its user: what the AI owes the user (honesty, loyalty, care), and what makes care non-negotiable as an architectural constraint rather than a setting.
Nicholas Templeman · MEOK AI LABS · 2026
Current AI memory systems discard temporal metadata, flattening all memories into a single atemporal store. We propose temporal memory chains: a linked-list architecture that preserves formation time, relational context, and significance trajectories — enabling AI to reason about how its understanding of a user has evolved.
Nicholas Templeman · MEOK AI LABS · 2026
We describe a cryptographic architecture for AI memory that ensures only the user can read, modify, or delete their AI's memory store. Server-side, all memory is encrypted under user-held keys. The AI processes memory through a secure enclave. No operator — including MEOK — can access user memory.
Explore further
The care-alignment contract between you and your AI. Read the full specification.
Read the Maternal CovenantFour live experiments: pgvector HNSW search, consciousness state machine, dream-state creativity, Byzantine council voting.
Explore MEOK LabsRead and run the Byzantine Council, memory schemas, and character engine yourself.
Browse the open source reposCare-aligned AI is an alignment approach that makes care non-negotiable — not a setting the user can turn off, but a constitutional requirement baked into the AI's architecture. The Maternal Covenant is MEOK's implementation: a formal, verifiable contract between an AI and its user.
BFT is a property of distributed systems where the system continues functioning even if some nodes fail or act maliciously. MEOK applies it to multi-agent AI governance: 33 agents must reach consensus before executing sensitive actions. No single agent can cause harm unilaterally.
Temporal memory chains are MEOK's approach to AI memory that preserves the timeline and emotional weight of memories — not just facts. Each memory is tagged with when it was formed, how it relates to others, and its significance score.
We actively seek research collaborators in AI alignment, governance, and privacy architecture. Email research@meok.ai with a brief description of your work and how it relates to sovereign AI.
MEOK AI LABS publishes open research in four areas: Byzantine fault-tolerant consensus for multi-agent AI governance (MEOK-AI-2026-001), the Maternal Covenant care-alignment framework (MEOK-AI-2026-002), temporal memory chains for AI memory with preserved context (MEOK-AI-2026-003), and sovereign memory architecture using cryptographic user-owned memory stores (MEOK-AI-2026-004). All papers are authored by Nicholas Templeman and published freely. MEOK AI LABS is a UK-based independent research lab.
We are actively seeking research collaborators in AI alignment, governance, and privacy architecture.
research@meok.aiOpen research · Free to read · Free to use