References that matter
Discuss selected images and documents against their actual content. Keep citations tied to the relevant source or page.
The new ZekeChat is being developed as the shared conversation experience inside Zeke’s Computer and in a standalone client. It is designed to be a useful project partner, with context and tools around the conversation.
Bring an unfinished idea, a draft, a screenshot or a difficult question. The intended experience starts by helping you think: offer a strong alternative, challenge an assumption, or ask a focused question when its answer materially affects the work.
Discuss selected images and documents against their actual content. Keep citations tied to the relevant source or page.
Recall approved project context, preserve the distinction between fact and preference, and support correction or forgetting.
Investigate current questions, retain source evidence and make uncertainty visible instead of turning an inference into a fact.
Carry scope and dependencies into the next action. Accept corrections, stop work when asked and resume against current state.
These capabilities are requirements of the new experience, not a blanket release claim. The available October implementation remains in progress. Real-model, hardware and end-to-end acceptance are separate from deterministic tests.
A focused place to talk to Zeke. The existing October local client and its conversation data are retained while the shared successor is developed.
The same product direction alongside files, project context and supported tools. The integration must reuse Computer’s authority and storage rather than create a second source of permissions.
Conversation history is private working material. Moving existing local history into cloud storage is a separate choice; a new client does not imply automatic upload or cross-device synchronization.
The new experience targets the maintained Bonsai 2/Hadamard stack. Computer remains a cloud-native workspace; the default design pairs it with an authorized local runner for inference. Cloud workspace does not mean the model is hosted or always awake.
The runner must be paired, reachable and allowed to use the selected inputs. If it is asleep, AI work waits. There is no implied paid fallback, and a ChatGPT subscription does not pay for another product’s runtime API calls.
Dotsys supplies the project and consent approach for ChatGPT Dots users. ZekeChat is a companion product with its own implementation. They can share an ecosystem direction without becoming the same assistant, license or runtime.
The current Dotsys operational license keeps the assistant role with ChatGPT Dots and supporting Codex workflows. This page does not grant permission to operate the licensed Dotsys materials through Zeke or another non-OpenAI assistant.