Local storage and online capabilities coexist
This article was first published in February 2026 about an early local-storage design. The November 2026 revision distinguishes that history from current accounts, model channels and tool services.
Desktop workspace files, conversations and some settings can stay on your device. Online inference can send the prompt, selected context or file excerpts needed for a task to the configured model service. Accounts, payments, enabled cloud features and third-party tools may involve separate processing paths.
Check actual features and destinations
- Your own API key: inspect the endpoint and provider policy, including any proxy at the configured URL.
- Tokeny channels: check account eligibility, channel details and the privacy policy.
- Local models: verify the inference endpoint; web search and remote MCP tools may still use the network.
- Local MCP processes: stdio describes communication, not an offline guarantee.
Protect local files
Back up workspaces and important documents. Close the application before migration and follow current version instructions. Deleting a local folder does not replace deleting server-side account data; use the channels in the privacy policy for account and data requests.
Test with a small sample
Use synthetic input to verify models, tools and output paths before real documents. Reading local history and completing an offline AI task are different capabilities. The latter requires the model, material and all necessary tools to be available locally.
Read the current model and data guide, the privacy policy and the getting started guide.