Membrain
Membrain is my workspace for documents and AI. I am building an interface to bring that work together.
Security objective
A standard above ordinary market practice
The programme’s ambition is to develop a data-handling and confidentiality system that is more demanding than ordinary market practice. This is a design objective that must be demonstrated through technical, security, privacy and operational testing; it is not a quality claimed as proven today.
What the project is intended to solve
What the project is intended to solve
Membrain is not being developed as a single chatbot or model. The objective is a configurable and governed working environment in which documents, organisational knowledge, processes, tools and execution using AI, deterministic software or both can form part of the same traceable workflow.
The architecture separates custody from processing, and processing from decision-making. A task can be routed to people, local tools, specialised agents or an authorised external model without giving one provider control over the whole system.
Conceptual architecture
Conceptual architecture
Sources and custody
Documents, email, databases and other organisational sources are intended to enter according to defined permissions. Originals and sensitive knowledge are intended to remain behind the Glaux boundary.
Knowledge and context
Structured documents, versions, metadata, indexes and semantic relationships are intended to provide the right amount of context for each task.
Orchestration and execution
Workflows are intended to allocate bounded tasks among people, local tools, specialised agents and authorised external models. Each participant should receive only the context required by its role and task.
Governance and decision
Roles, policies, budgets, activity records, quality controls, stop mechanisms and approval points are intended to keep the system under human responsibility.
Design principles
Design principles
- Architecture before model
- Process, data flow, responsibility and control are designed before selecting the model that will perform a task.
- Local where possible
- Local processing is preferred when it can solve the task; external capacity is used only when justified and authorised.
- Replaceable providers
- Models and services should remain replaceable so organisational knowledge and working methods are not locked to one provider.
- Observable human control
- Activity should be monitorable, stoppable and auditable; actions with external effects require appropriate human approval.
Role in the wider system
Role in the wider system
Membrain is intended to become the progressive integration point for Glaux and Alonides. Mnemon may then contribute the relational, semantic and ontological structure. This is the design direction, not a claim that full integration already exists.
External models can add capacity, but they should not automatically receive the full context, control original material or serve as the final decision authority.
Areas of work
Areas of work
- 01
Configurable workflows for documents, matters and organisational knowledge
- 02
Automation using AI, deterministic software or a controlled combination
- 03
Roles, approvals, traceability and human oversight