What is a local-first AI workspace?
A plain definition, what "local-first" does and does not mean for AI tools, and five questions to ask any product that claims it.
A local-first AI workspace is an app that keeps your projects, files and conversation history on your own computer by default, and brings the AI assistants and tools you use into one place to work on them. "Local-first" describes where your work lives. "Workspace" describes how it is organised: around projects, with AI helping inside them.
The term is new to AI products, and it is easy to stretch, so this page sets out what it should mean and how to check a claim.
Where "local-first" comes from
The idea comes from an influential 2019 essay by the research lab Ink & Switch, Local-first software: You own your data, in spite of the cloud. It argued that software should keep the primary copy of your work on your own device, work without waiting on a server, sync between devices when it can, and leave you in control of your data for as long as you want it.
Those principles were written for documents and design files. AI raises a new question, because the most capable AI models usually run on someone else's servers.
What local-first means for AI tools
For an AI workspace, local-first is a promise about defaults:
- Your work is stored on your computer. Projects, files, notes and the history of your conversations live on your machine first.
- Some intelligence runs on your computer. Smaller models can handle tasks like searching your own files, sorting notes or drafting short replies without anything leaving your machine.
- Sending work out is a visible choice. When a task needs a larger model from an AI provider, the part of your work that model needs is sent to that provider. A good local-first tool makes it clear when that happens and lets you decide.
What it doesn't mean
Local-first does not have to mean offline. A local-first workspace can still use cloud AI models, and many tasks will need them. The difference is that your work's home is your computer, and the cloud is a place you send things to on purpose.
It also doesn't mean "nothing ever leaves your computer". Any product that uses a cloud AI model sends that model some of your content. Be wary of anyone who claims otherwise while also offering the latest cloud models.
Five questions to ask any tool that claims it
- Where are my projects and history stored by default? The answer should be "on your computer", with sync as an option.
- What leaves my computer, and when? You should be able to see which content goes to which AI provider, task by task.
- Can I set permissions for each tool? Connecting your email should not mean an assistant can read your calendar, and vice versa.
- Does anything act on my apps without my approval? Sending, posting, deleting and paying should wait for you.
- Can I take my work with me? You should be able to export your projects in formats other software can read.
A product that answers all five clearly has thought about this. One that answers with general reassurance probably hasn't.
How Pelagos approaches it
Pelagos is the local-first AI workspace we are building. It runs on your Mac, uses local intelligence where it can, and organises your work into projects and rooms. You set permissions for each connected tool, and anything that acts on your apps waits for your approval. When you choose to use AI workers from other providers, the work you send them goes to those providers.
Pelagos is in private preview and Mac first. You can join the waitlist below.