The foundation of your personal AI
More possibility. Still your control.
The foundation that brings your models, tools and permissions into one personal AI experience.
The best model for thinking may not be the one you want doing the work. We’re building Neel to bring those choices together, with your permissions, your context and a clear view of what you share. The technology underneath should earn its place through the experience it gives you.
What matters to you
One system. Built around your choices.
This is what we’re building toward. Every Neel surface runs on this engine and launches together, on every platform.
Choose intelligence, not a silo.
Use a compatible local model for private work or opt into a supported provider. The direction is a workflow where you choose the right model for each task, without rebuilding your setup around it.
Stay in control of the work.
Set what your agent can read and change. Inspect the requests, responses and tool actions behind an answer, so handing over a task does not mean handing over blind trust.
Keep your work with you.
We’re working toward encrypted communication between paired devices: continue a task, follow its progress and approve actions from your phone. Cross-device capabilities are in development.
Build a system around your life.
Your instructions, preferred models, tools and permissions should work together. The ambition is one personal setup across your workspace, terminal and mobile—not a collection of disconnected chats.
Performance, with evidence
Less waiting. More work done.
That is the goal. Our public benchmark report is not available yet; the activity illustrations are not speed claims.
Time to first response
How long you wait before a response starts, with cold starts and already-loaded models measured separately.
Generation speed
Tokens per second, alongside the exact model, runtime, hardware and context length. A number without its setup tells only part of the story.
Time to a completed task
The whole workflow: model calls, tools, permissions and verified outcomes—not just how quickly text appears.
We will publish results with their test setup, task and limitations. Local performance depends on your model and hardware; hosted performance also depends on the provider and network.
Confidence through visibility
See what happens behind the answer.
Follow model calls and approved tools, inspect the exact exchange and compare the raw response with its presentation.
Early access
Make room for your own AI system.
Tell us how you want your models and devices to work together. Join the waitlist for early-access updates.