What does "local AI" actually mean?
Most AI assistants send everything you type, and everything they read from your tools, to remote servers for processing. Zenpa doesn't. The AI model is downloaded once and runs entirely on your device, using its own processing power. When Zenpa reads a Slack thread, summarizes an email, or answers a question about a project, that computation happens on your machine and nowhere else.
This is what we mean by local-first AI: the model, your memory, the search index, and every answer Zenpa gives you live on your device, under your control.
What stays on your device
- The AI model, which runs offline and on-device. By default, no cloud AI provider is involved at any step.
- Your memory, everything Zenpa learns from Slack, Gmail, Jira, Notion, Finder, and your other tools is stored in an encrypted database on your device.
- Your answers, every response Zenpa gives you is generated and kept locally.
- Your files and messages, Zenpa pulls your latest memories from the tools you already use without uploading their content anywhere.
Want a bigger model for certain answers? Cloud Boost is an optional setting that connects a cloud model using your own API key. It is off by default, and when you use it, only the memories relevant to your question are sent, nothing more.
The only things that touch the network
Zenpa uses a secure connection for a short list of things: verifying your account when you log in, connecting your tools, usage telemetry that helps us improve Zenpa (you can turn it off in Settings), and feedback you choose to send. Your memory, files, and answers are never part of any of these. Beyond that, Zenpa works offline. You can close your laptop, get on a plane, and Zenpa keeps answering questions about your work, because everything it needs is already on your device.
Why local beats cloud for an AI that sees your work
An assistant is only useful if it can see your real work: your messages, decisions, files, and conversations. Some data should never be put in the cloud, by design. With on-device processing, there is no server to breach, no training pipeline ingesting your documents, and no vendor quietly mining your company's knowledge.
- Your data can't leak from a cloud you're not in.
- No third party can be subpoenaed for data it never had.
- Latency is your hardware, not a round trip to a data center.
- It keeps working when the internet doesn't.
You stay in control of the memory itself, too: pause it, exclude specific tools, or clear anything, anytime. Read more about how we secure your data on our Security page, or why we built it this way in our Manifesto.