Keep the understanding.
Your conversations end. The understanding doesn’t have to. Mnemara manages the preferences, decisions and experiences that help AI understand you over time.
Independent memory. Owner-controlled.
Own your
memory.
Take control of what your AI knows about you.
Mnemara is being built as a secure, independent long-term memory layer. Keep your preferences, decisions and personal context under your control—and choose which supported AI tools may use them.
Working prototype: ChatGPT + Mac.
Broader integrations planned.
Useful memory. Uneasy ownership.
How much of yourself do you want one AI company to hold?
Your relationships. Your finances. Your private doubts. At work, your plans, customer information and intellectual property.
The more useful an assistant becomes, the more it learns. You may trust its provider today—but should changing assistants mean leaving that understanding behind, or rebuilding it from scratch?
You should not have to choose between useful AI and control over your information.
Memory in action
It remembers.
You decide.
A fresh conversation. A useful memory. A private detail that needs your say-so.
Read the video description and transcript
Ask naturally. In a fresh ChatGPT conversation, the user asks, “How is my Adobe investment going?” The question does not supply the purchase details.
You decide. A native Mnemara notification says “Approve a memory request.” The user opens Options and selects “Allow once”. ChatGPT explains that the private purchase details require approval; the user types “i have approved the request”.
Put memory to work. ChatGPT says it has retrieved the Adobe investment details from Mnemara. Using the example holding of 100 shares purchased at US$150 and the recorded quote of US$232.77, it shows a value of US$23,277 and an unrealised gain of US$8,277 before fees and currency effects.
Own your memory. Take control of what your AI knows about you.
This demonstrates memory access and owner approval—not investment advice or independent verification of market data. General market commentary and waits were trimmed. No app screens or approval actions were fabricated.
Remembering something and sharing it are two different decisions.
Why independent memory matters
Your memory. Your choices.
Control access.
Keep private memory without giving every connected AI unrestricted access. Mnemara checks requests against its policy and asks for approval when required.
Keep your freedom.
We’re building towards interoperability across devices, AI vendors, agent harnesses and personal assistants. Your memory follows you; you decide what each supported tool may use.
Working together
Not another assistant.
Not another notes app.
Your assistant helps you do things. Your project files hold the detailed work. Mnemara manages what your AI remembers about you.
Project knowledge
The presentation deck in your files or Obsidian vault.
Remembered experience
“Last time the deck had too much text. Lead with the demonstration.”
Personal profile
“I prefer direct, practical explanations.”
These describe different purposes, not exclusive storage locations. Obsidian can hold personal knowledge too; a Mnemara connector is not available today.
Mnemara does not replace your assistant or your project files. It gives you control over the memory that connects your conversations, decisions and work.
Permission, not assumption
Kept in memory doesn’t mean shared with every AI.
The AI can suggest what to remember and ask for what it needs. Mnemara applies its own rules before storing or releasing information.
Your AI asksFor a saved memory.
Mnemara checksPolicy and permission determine the next step.
Permitted → shareReturn the relevant memory.
Approval needed → ask youShare only if you approve.
Not permitted → withholdDo not share the memory.
Security needs substance
Trust needs more than a promise.
Our starting point is an encrypted vault on your Mac, with permission checks separate from the AI asking to use it.
Understand the limitations →- Encrypted local storage in the current Mac prototype.
- Separate checks for saving and disclosing memory.
- Owner approval for protected requests.
- Visible memories, revisions and activity.
- Local policy checks that do not simply trust the AI’s classification.
Approved information is sent to the requesting AI. Mnemara cannot control how that provider handles it afterwards. A working prototype is not a claim of independently verified, production-grade security.
A working start
Starting with people.
Built with a broader purpose.
For people who already find AI useful—and want more say over what it knows.
Today
A working ChatGPT-and-Mac prototype: fresh-chat recall, sensitive disclosure approval, approved saves and corrections, native notifications, and a local owner app. Your Mac must be online.
Some controls in the prototype are previews. See the capability notes for what is demonstrated and what is still planned.
Next
A limited alpha with easier installation, stronger recall and dependable owner controls. Broader integrations, phone approvals and optional cloud availability are planned—not available today.
Teams, enterprises and government/sovereign deployments are longer-term directions.
Keep what matters
Change your tools.
Keep what matters.
Memory should travel with its owner.
An email introduction, not an instant download. We’re preparing a small, supported alpha.