Your intelligence layer
One AI.Every model.One memory. Always yours.
Kelven chooses the right intelligence for every task while keeping your memory, preferences, projects, and privacy under your control — no matter which model or AI interface you use.
KelvenCompare these two business ideas and tell me which one has better economics.
- Understanding task
- Retrieving relevant context
- Selecting intelligence
- Generating answer
The subscription idea has the stronger economics. It compounds: each customer keeps paying while acquisition cost stays fixed, so margin improves over time.
- · Recurring revenue instead of one-off transactions
- · Roughly 3x higher gross margin at the same volume
- · The marketplace idea only works with heavy supply investment first
If speed to first revenue matters more than long-term margin, the marketplace is the shorter path — but it plateaus.
The intelligence layer
One question. Kelven handles the rest.
Different AI models are better at different things. Some reason better. Some write better. Some research current information better. Some are faster, cheaper, or better suited for private work.
Kelven understands what you’re trying to accomplish, applies your memory and privacy preferences, and chooses how the task should be handled — so you don’t have to keep comparing models yourself.
- Your request
- Kelven
- Understand the task
- Retrieve relevant memory
- Apply privacy rules
- Choose the right intelligence
- Answer
Sending every question to the largest model available is slow, expensive, and burns energy for no gain. Sending it to the wrong one gives you a worse answer.
Kelven matches the task to the intelligence that actually suits it.
Simple questions come back faster on efficient models. Hard ones get the heavier reasoning they need. You get a better answer, sooner, with far less compute — and a smaller energy footprint behind every one.
Privacy comes along for the ride: Kelven sends only the context a task needs, so there is nothing to trade away.
- Reasoning
- Research
- Coding
- Vision
- Tools
Light path. Fast, direct.
Routing
The best model isn’t
always the same model.
A coding problem wants one kind of intelligence. Deep research wants another. A sensitive document may call for a route with different privacy characteristics. A simple question may not need an expensive frontier model at all.
So the question isn’t which model is best. It’s which model is best within your rules.
Quality · Reasoning · Speed · Cost · Privacy · Data sensitivity · Tool availability
Weighting
- Quality
- Speed
- Cost
- Privacy
Good answers without reaching for the most expensive route every time.
Candidate routes
- Frontier reasoningDeepest reasoning, highest cost
- Long contextSelectedHolds a whole document at once
- Privacy-configured routeReduced retention, narrower context
- Fast generalQuick answers for simple work
- On-deviceNot selected — task exceeds local capability
Illustration of Kelven’s intended routing model. Routing controls are in development and are not available yet.
Portable memory
Your AI shouldn’t forget you when you switch models.
Most AI products build their own isolated memory of you. Switch platforms and you often start over.
Kelven separates your memory from the model underneath it. Your preferences, projects, decisions, working style, and context can remain part of your intelligence layer even as the AI you use changes.
One memory. Always yours.
Models change. Interfaces change. Your context stays with you.
What Kelven memory can contain
- Your preferences
- Your projects
- Your decisions
- Your working style
- Your recurring instructions
- Your important relationships and context
- Your privacy preferences
Kept intentionally, not collected exhaustively. You can see every item, edit it, delete it, and decide whether it may ever be shared outside Kelven.
Memory
Your memory.
Your choice.
Most AI products make memory a platform decision. Kelven is being built to make it yours — one control, three honest answers.
Memory is stored by Kelven. Relevant context still travels to the AI provider handling a request.
All three are available in Kelven today, chosen from the app and applied to every conversation until you change it.
Privacy by design
Remember more.
Share less.
Kelven can know what matters to you without sending everything it knows to every AI model.
For each request, Kelven is designed to retrieve only the context that matters, apply your privacy rules, and remove or tokenize unnecessary sensitive information before sharing it externally.
Your memory belongs to you. Models only get what they need.
Kelven context, on your device
Jordan from Northstar Labs needs to respond to Casey about Project Aurora.
What an external model may receive
PERSON_1 from COMPANY_1 needs to respond to CLIENT_1 about PROJECT_1.
Let AI understand the problem without necessarily knowing everything about you. The mapping back to real names stays on your device.
Kelven chooses the intelligence — and what that intelligence gets to know.
This is minimisation, not invisibility: an external provider still processes the abstracted request. Credentials — passwords, API keys, recovery phrases — are removed outright rather than replaced with anything reversible.
Memory used for this answer
3 memories · 1 file · no new memory stored
- ViewRemoveAnswer again without itNever use this
Personalisation you can see. This is the transparency Kelven is being designed around, not a shipped feature.
Transparency
You should be able to see
why an answer knows you.
Invisible personalisation is the part of AI people find hardest to trust. Kelven is being designed so the personal context behind an answer is visible next to it — and removable from there.
Model independence
The best AI won’t always be the same AI.
Today it might be GPT. Tomorrow it might be Gemini, a model running on your own device, or something that hasn’t been released yet.
Kelven is designed so your AI experience doesn’t depend on whichever company happens to have the best model today.
Models can change.
- Your memory stays.
- Your preferences stay.
- Your projects stay.
- Your privacy rules stay.
Your intelligence layer stays with you.
Your choice
Cloud when you want it.
Local when you don’t.
The same Kelven — same memory, same privacy rules — whether the thinking happens in the cloud or on your own machine.
Cloud when you want it
Kelven reaches GPT and Gemini models, and picks between them on task fit — speed, reasoning, context and your privacy rules.
Local when you don’t
Choose Local Only and the work runs on a model installed on your own device, so nothing leaves it.
Never a silent switch
If a local model can’t do the job, Kelven stops and asks — it never quietly sends the task to the cloud instead.
Running on your device takes a one-time model download and a recent computer or phone with graphics acceleration. On-device models are smaller than the cloud ones, so they are quicker and completely private, but less capable on hard work — Kelven tells you which you are using, every time.
Interface independence
Use Kelven — or bring Kelven with you.
The Kelven app gives you the complete experience: intelligent routing, portable memory, privacy controls, projects, and model selection.
But Kelven isn’t designed to become another closed AI silo. Through compatible integrations, selected Kelven memory, context, projects, and preferences can travel with you into AI interfaces you already prefer.
Your interface can change. Your intelligence layer doesn’t have to.
Kelven app
Complete experience
Claude
In testing
ChatGPT
Experimental
Other MCP clients
Compatible
The intelligence layer
Kelven
Routing, memory and privacy rules live here — not inside any one interface.
Kelven controls what Kelven shares. It can’t hide something you typed directly into another app.
Complete experience
The Kelven app
Chat, files, projects, memory, routing controls, privacy modes and per-answer receipts. Installable on a phone home screen.
In testing
Claude
Connect Claude to the MCP Gateway and it can ask for the narrow context a task needs — after you approve the permissions.
Experimental
ChatGPT
Where MCP connections are available, the same Gateway applies. We won't call it supported until we've verified it ourselves.
Compatible
Other MCP clients
The Gateway is a standards-compliant MCP server, not a per-host integration. Compatible clients work the same way.
Interface independence
Kelven goes where you work.
The Kelven MCP Gateway lets compatible AI applications request selected Kelven context without receiving unrestricted access to your entire memory or profile.
One Kelven connection. Compatible AI interfaces.
Relevant memory
Only the memory items a specific task actually needs.
Project context
A summary of the project being worked on, not every project.
User preferences
How you like to work, write and receive answers.
Privacy preferences
The rules that decide what may be shared at all.
Selected tools
A small set of read-only requests, not open access.
We describe interfaces as compatible MCP clients until we have tested them ourselves. Kelven also can’t conceal something you typed directly into another app —Kelven controls what Kelven shares.
The product
One way to ask. Any kind of task.
You don’t pick a tool or a model — you describe what you need.
Ask
Anything, in one thread that keeps its context.
What changed in my plan this week?
Three things — here they are.
Research
Sourced answers, gathered and synthesised.
Market outlook, 2026
12 sources · summarised
Create
Drafts written the way you write.
Launch note — draft 2
Concise, direct, no filler
Analyze
Documents and numbers, read carefully.
Agreement.pdf · 70 pages
6 risks identified
Code
Build, review and explain.
api/router.ts
Refactor suggested
Plan
Turn intent into a sequence of steps.
Q1 rollout
4 milestones
The app
One place for everything.
Chat, files, projects, memory, routing controls and per-answer receipts.
Should we expand into Europe this year?
- Context3 relevant memories
- ResearchCurrent information
- ReasoningDeep analysis
- IntelligenceSelected automatically
Yes — but not as a full market entry this year.
Your margin structure only works above roughly 40% gross, and localisation plus support coverage would take you below that until volume catches up. A single-country pilot keeps the downside contained.
- Pilot one country, keep pricing in EUR only.
- Revisit full entry once support cost per account drops.
How it works
Five steps, every time.
- 01
Understand
Kelven determines what you’re trying to accomplish.
- 02
Remember
Kelven retrieves only the memory and project context relevant to the task.
- 03
Protect
Kelven applies your privacy rules and minimizes what information gets shared.
- 04
Route
Kelven chooses the appropriate model, intelligence, or tools.
- 05
Respond
You receive the result without manually deciding which AI to use.
Trust
Memory should create trust, not take it away.
Kelven is being designed so people remain in control of what their AI knows about them.
Transparent
See the context Kelven remembers.
Controllable
Edit or remove saved information.
Relevant
Use memory only when it helps answer the current request.
Access
One subscription.
One intelligence layer.
Instead of managing several AI subscriptions and deciding which one to open, Kelven is one intelligence layer across the AI ecosystem.
Baseline privacy controls are included on every plan, including Free — never paywalled.
Free$0
The core experience, with privacy controls included.
Pro$25 / month
For people who use AI every day.
Power$60 / month
For people who push AI harder.
EnterpriseFrom $4,500 / month equivalent
Up to 50 employees, billed annually, plus AI usage.
For builders
Built for people.
Available for builders.
The same intelligence layer behind Kelven will eventually be available to developers. Send a task instead of hard-coding a model. Kelven determines how it should be handled.
const response = await brand.generate({
task: "Analyze this document and identify the key risks."
});Enterprise
One intelligence layer for your organization.
Enterprise starts at $4,500/month. Includes up to 50 employees. Additional seats are $60 per employee per month. Billed annually. AI usage is additional.
- Centralised AI routing
- Shared organisational context
- Privacy and provider rules
- Usage and efficiency visibility
- MCP Gateway access
- Admin and billing controls
The right AI, automatically
Kelven matches each task to the intelligence that suits it — never to a brand.
Privacy before the cloud
Sensitive details are found and protected on your device, before anything is sent.
One memory. Always yours.
Your context isn’t trapped inside one AI company.
Less wasted intelligence
Simple questions don’t need the heaviest model, so answers come back sooner.
Cloud or local — your choice
Run a task in the cloud, or keep it entirely on your own device.
Your intelligence layer
Your AI will change.
Your memory shouldn’t.
Use the best intelligence available without rebuilding your context every time the AI landscape changes.
The future isn’t another AI model. It’s an intelligence layer that’s yours.
Kelven is the user-controlled intelligence layer between you and AI — with one memory that’s always yours.
No spam. Just one note when early access opens.