Guide
What a local-first AI workspace is
A local-first AI workspace is a place on your computer where AI work lives. Compatible local models run on your machine. Supported providers enter only when you send a turn out. The project, files, and approved memory stay with you when intelligence changes.
The workspace is the product
Most AI products are a chat owned by a provider. You type, they store, you lose the thread when you change tools. A local-first AI workspace inverts that. The project is the durable object. Models are interchangeable.
Velociti puts compatible local models, supported providers, files, instructions, and agents in one operating layer. Changing intelligence does not mean starting over.
What stays on your computer
Prompts, context, responses, and thinking stay on your machine. Search and reopen them with the project. Polar retrieves relevant file segments without injecting the entire workspace.
A supported provider is used only when you send a turn out. Work is not used as training data.
Local models and your keys
Install a compatible local runtime, measure hardware, and add the model to your picker. GGUF weights can sit in system RAM, including DDR5, as well as GPU VRAM.
Connect ChatGPT, Claude, Grok, and the rest through keys you already control. The same project can move from a local model to a supported provider without reloading files.
Questions
- Is a local-first workspace the same as running Ollama?
A runtime loads weights. A workspace keeps the project: files, instructions, approved memory, receipts, and the next turn. See Velociti vs Ollama.
- Do I have to stay offline?
No. Local is the default. A supported provider enters when you send a turn out.