DOCUMENTATION / Configuration
Choose an execution and connection path
Connecting ChatGPT
Hosted agent access is in setup and is not available yet. Solar System remains available for ordinary browser exploration without login. A private ChatGPT pilot will be released only after authenticated browser operation and disconnect have been verified. A custom pilot connection is not a public marketplace listing.
The intended setup is: open Solar System, choose Agent access, sign in and explicitly enable read access or control for that tab. Add the VizualModel HTTPS MCP connection in ChatGPT once, using the OAuth configuration shown by ChatGPT, and sign in with the same identity. In the conversation, list sessions and choose the tab by its human-readable label. Every operation targets that session explicitly.
Do not configure a connection from an unverified endpoint or paste credentials into a conversation. Exact ChatGPT plan support, callback URI and tested interface will be recorded when the real pilot connection is verified.
If no session is ready, open the model and enable access. If the tab is asleep or reconnecting, bring it to the foreground and wait for its status. If access was revoked, enable it again deliberately. An incompatible build requires the supported model version. When several tabs are listed, choose one explicitly.
Connecting Claude
Host-specific instructions pending verification. A suitable local MCP host may operate the documented stdio adapter, but this phase does not establish tested Claude compatibility. Consult the model’s versioned MCP documentation and the host’s current documentation.
Connecting Goose
Host-specific instructions pending verification. The baseline includes local MCP operations, but Phase 2 does not certify a Goose connection flow. Check the adapter, ready-browser requirement and permissions in the repository guide.
Connecting Codex
A local MCP observation was verified during Phase 1 through the existing environment. A portable, independently tested Codex setup recipe is still pending. Use the versioned MCP guide as the implementation reference; do not infer remote-host support.
Local versus hosted execution
The browser application contains the model, calculation and visualization. The proposed hosted service only authenticates users and routes temporary commands and results between ChatGPT and the approved tab. ChatGPT remains the conversation interface; the model does not read or embed your conversation. Keep the tab open and responsive; background and mobile tabs may be suspended.
Developer setup: the existing local MCP bridge remains documented in the versioned local service guide. It requires Node.js, a local service and per-run credentials. This is a developer path, separate from the intended hosted customer journey. Local model execution does not mean agent inference is local.
Authentication, permissions and security
Keep credentials and connection records private. The hosted release requires one verified identity across browser sign-in and ChatGPT OAuth, separate read/control permission, explicit per-tab consent and revocation. A session ID identifies a tab; it does not grant access. Hosted authentication and session isolation remain unverified until pilot acceptance passes.
Adapting model source requires access to the source, permission under its license and an agent environment that can edit and run it. Availability through a hosted application alone does not provide these capabilities.