Opaque vendor custody
It is often unclear what has been remembered, where it came from, how it is used or how to correct and remove it.
For investors
Long-term memory is becoming the control point for AI.
AI is becoming more trusted with more of our lives—while people and organisations are losing trust in AI vendors to manage their information.
Mnemara is building the owner-controlled memory authority beneath the AI tools people use.
Start an investment conversationThe problem
The more useful AI becomes, the more deeply it knows us.
That knowledge is not just chat history. It is durable memory: personal context, operating decisions and institutional knowledge that increasingly sits inside opaque vendor profiles.
Deeply sensitive information
Memory may include health or psychiatric information, sexual preferences, finances, family details, corporate secrets, intellectual property, or citizen data such as identities and tax records.
Lock-in is structural
Memory is what makes AI work efficiently. The more useful the context, the harder it is to change vendor without making the next AI relearn everything.
The answer
Memory should be controlled by its owner, not by the AI vendor.
Mnemara is an owner-controlled memory layer, separate from the model vendor. It can live locally on a device or in a cloud environment the owner chooses. AI tools borrow only relevant, approved context; they do not become the default owner of the vault.
Owner-controlled vaultLocal-first memory or a chosen managed environment.
Policy and permissionControl what is stored, shared, approved, paused or never stored.
Relevant context onlyRelease the small slice an AI needs for the task.
Portable by designCarry approved memory across supported AI tools.
Why now
AI use is spreading faster than memory ownership.
People already move between ChatGPT, Claude, browsers, coding agents, internal tools and specialised assistants. Each tool is becoming better at collecting context. There is not yet a widely adopted owner-controlled layer for carrying that context safely between them.
Not another note app
Tools such as Obsidian are excellent human-owned project libraries: notes, files, research and links. Mnemara complements them: project files are the evidence; governed memory carries forward the decisions, constraints and context an AI needs.
Similar projects prove the need
Many memory projects focus on retrieval quality, efficiency, developer infrastructure or agent capability. Mnemara focuses on custody, permission, audit and portability.
The trust problem is immediate
Individuals worry about their private life; organisations worry about IP and operating knowledge; public institutions worry about protected and citizen information.
Working proof
This is more than a thesis.
The current private-preview prototype has demonstrated the core local-first path with ChatGPT: a paired local companion, an encrypted synthetic vault, approval-gated changes, auditability and controlled retrieval through a standard MCP connection.
What has been proven
Natural-language retrieval, approval-gated writes and corrections, revisions/audit, offline truthfulness and paired remote access to a local synthetic vault.
What is deliberately not claimed
This is not a finished consumer product, a compliance certification or a claim that every AI tool will route memory perfectly without configuration.
Why it matters
The architectural boundary is real: the model borrows a relevant slice of context; it does not become the default owner of the vault.
Roadmap
A focused path from proof to platform.
V1 · Consumer CompanionLocal memory, clear onboarding, supported AI connection, approval and audit.
V2 · Portable continuityCloud backup/sync, multi-device use and premium cross-tool portability.
V3 · Enterprise memoryShared project knowledge, policy, permissions, administration and governed deployment.
V4 · Sovereign deploymentPrivate-hosted, jurisdiction-controlled memory, implementation and support.
Consumer adoption is the product-learning wedge. Enterprise and sovereign deployment are deliberate expansion paths—not near-term assumptions.
Commercial path
Free ownership. Paid convenience. Governed deployment.
Consumer
Free local memory, with paid cloud backup/sync and premium portability across supported AI tools.
Enterprise
Commercial model to validate: managed workspace memory, governance, policy, audit and deployment support.
Government and sovereign
Commercial model to validate: private-hosted deployment, implementation, support and jurisdiction-specific controls.
Funding would accelerate the consumer product, build governed enterprise and sovereign deployment options, and establish the sales and channel capability required to bring those products to market.
Talk with Mnemara