Who is acting?
A ChatGPT Dot, Codex, Zeke, a person, or a connected service. Their identities and authority are distinct.
RUNTIME & AUTHORITY
Know who is acting.
Know where it runs.
The ecosystem combines a System approach, OpenAI products and independent companion software. A shared brand or architecture diagram does not make their accounts, permissions, storage or runtime the same.
THE ROLE MAP
One picture. Distinct boundaries.
| Part | Where / role | Scope and status |
|---|---|---|
| ChatGPT Dots | OpenAI’s ChatGPT environment and Dot computer. | Intended operational assistant for licensed Dotsys adoption; actual capabilities depend on the host. |
| Dotsys System | Owner-controlled private Git repository and explicitly chosen deployments. | Project structure, documents and consent approach; not a hosted AI service. |
| Surface | A separately configured protected application. | Project overview and bounded operator console; not supplied as a runtime by the protocol package. |
| OpenAI Codex | An independently authorized Codex environment. | Separately approved implementation; repository and environment access must both be verified. |
| ZekeChat | Standalone client and Chat inside Zeke’s Computer. | New shared conversation experience in development; separate from the ChatGPT Dot. |
| Zeke’s Computer | Cloud-native browser workspace with its own connected services. | WebLinux implementation direction, not the Dot’s cloud computer. |
| Bonsai inference runner | An explicitly paired, permitted and available local runner in the default design. | Native runtime and hardware support must be checked; sleeping hardware cannot execute queued AI work. |
| Linux / browser jobs | A supported separately configured execution backend. | Not a capability created by a static website, manifest or prompt. |
| Clock | Configured schedule and durable occurrence/inbox facilities. | A reminder and its requested AI work have different availability requirements. |
| Public card / hello | Owner-approved HTTPS publication and human-mediated exchange. | Protocol 0.1.0; no identity attestation, automatic messaging or execution authority. |
THE QUESTIONS THAT MATTER
Every action should have
an understandable answer.
What did I allow?
Identify the input, current scope and destination. Capture, build, publication, spending and access are separable.
What is available?
A configured interface is different from a reachable runner. State the real capability and the evidence for it.
What came back?
Keep the candidate, verification result and accepted revision distinct. Preserve partials and uncertainty.
COST & CONTINUITY
Cloud-native does not mean
always-on inference.
The Computer design can keep project state and scheduled reminders separate from model availability. AI work still needs a permitted, awake runner or a genuinely provided hosted route.
Local inference still has hardware, storage and electricity requirements. External APIs and hosting have their own eligibility and costs. No new paid model, provider entitlement or public exposure follows from adopting a prompt.