Guide
Switch AI models without losing the project
You should be able to change the model and keep the work. Files, instructions, and approved memory stay attached to the project. The next model reads what you approved, not a trapped provider transcript.
Why chats do not survive a switch
Provider chat accounts own the thread. Move from ChatGPT to Claude and you rebuild context by hand. Local runners have the opposite problem: they load a model and forget the job.
Velociti treats the project as the shared object. One click changes intelligence. The campaign file does not reload.
What actually stays
Files, instructions, approved memory, and tool gates stay with the project. A full transcript is not handed to the next model. Polar selects relevant file segments for the turn.
Receipts stay visible. A local step shows no provider charge. A provider step shows input and output you can inspect.
Local to provider, same desk
Run a compatible local model for the cheap, private pass. Send a harder turn through a supported provider with a key you already control.
That is the difference between a chat tab and a local-first AI workspace.
Questions
- Does the new model see every prior chat?
No. Approved memory and selected file segments travel. A full transcript does not.
- Can I switch from Codex to Kimi in one click?
Yes, when both are available in your picker. The project stays attached.