I would love to see a native Remalt desktop application for macOS and Windows, designed around local-first privacy.
The main reason is simple: as Remalt becomes more useful for RAG, personal knowledge, research, documents, MCP and AI workflows, it will inevitably contain some of a user’s most valuable and sensitive information.
I would like the option to keep that knowledge on my own computer rather than storing everything in the cloud.
Local-first Second Brain
Ideally, the desktop version could run the proposed My Second Brain / Personal RAG locally.
This could include:
• Documents and PDFs
• Research and captured webpages
• Personal notes and knowledge
• Writing and project history
• Embeddings/vector database
• RAG indexes
• Prompts and workflows
• API credentials
• MCP configurations
• Private business information
The user should be able to choose where this data is stored.
For highly sensitive work, there should be a Local Only mode where the knowledge base never leaves the device unless the user explicitly sends selected context to an external AI provider.
Clear privacy controls
I would like Remalt to distinguish clearly between:
Local data → remains on my device
Synced data → encrypted and synchronised between my devices
External AI data → explicitly sent to a selected model/provider
This would make it much easier to understand exactly where information is going.
Privacy should not depend on users simply trusting that their data is safe. The product should make data movement visible and controllable.
Bring Your Own API Keys
The desktop app could also support BYOK for providers such as OpenAI, Anthropic, Google and others.
My API keys could be stored securely on the device using the operating system’s secure credential storage.
This would allow Remalt to act as the interface and knowledge layer while users retain control over which AI provider receives each request.
Local AI models
Longer term, it would be valuable to support local models as well.
For example, users could connect Remalt to locally running models through tools such as Ollama or other compatible local inference systems.
That could enable workflows where:
Documents → embeddings → retrieval → model → output
all remain on the user’s computer.
Cloud AI could then become optional rather than mandatory.
Local MCP
The desktop application could also act as a secure MCP host/client.
My local Second Brain could expose selected tools and knowledge to authorised applications without making the entire knowledge base publicly accessible.
Permissions should be granular:
• Read knowledge
• Search/retrieve
• Add knowledge
• Modify knowledge
• Access specific collections
• Access specific projects
Each MCP connection should be visible, revocable and auditable.
Chrome Extension + Desktop
This would also work extremely well with the Remalt Chrome extension.
The browser becomes the capture layer:
Browse → Capture → Send to Remalt Desktop → Process locally → Index → Personal RAG
Instead of every captured article, webpage, PDF or note first being uploaded to a cloud service, the extension could send it directly to the local Remalt application.
That would make the combination of Chrome Extension + Desktop App + Personal RAG + MCP particularly powerful.
Why this matters
A Second Brain becomes more valuable as it learns more about the user.
But that creates a problem.
The more useful it becomes, the more sensitive it becomes.
It may eventually contain years of research, business knowledge, documents, ideas, decisions, writing, client information and personal context.
For that reason, I would strongly prefer Remalt to offer a local-first desktop architecture where the user owns the knowledge base and decides what leaves the device.
Cloud sync can still be available for people who want convenience.
Cloud AI can still be available when users want the strongest models.
But neither should be required simply to use your own knowledge.
For me, that would make Remalt considerably more valuable:
Chrome captures my knowledge.
Desktop keeps it private.
RAG makes it useful.
MCP connects it to my tools.
AI uses only the context I choose to share.
That is the kind of Second Brain I would trust with serious work.
