Deferred: optional Codex App Server backend with per-account isolation

Status

Deferred at the owner's request. Do not implement now. Current priority is a sparse, reliable Micro experience and functional app creation.

Proposal

Explore Codex App Server as an optional execution backend while Micro retains accounts, scoped services, inbox and delivery across clients.

Each user connects their own ChatGPT account. Isolate runtime execution, credentials, configuration, workspace and conversation state per tenant. Derive ownership from authenticated server identity, never model arguments. Scope every Micro tool call to that owner and the granted capabilities.

Persist credentials and conversation mappings across restarts; use supported managed authentication and refresh, with explicit reconnect on revocation. Start workers on demand and enforce per-account concurrency and usage limits.

Questions before implementation

• Confirm supported hosted multi-user integration, model availability and grant/subscription terms. Do not assume one open-source grant can fund all tenants or provides Astra Light.
• Evaluate production support: current official documentation labels app-server command and WebSocket transport experimental and unsupported for production workloads.
• Establish OS/container isolation, protected credential storage, tenant-bound tool access and recovery without replaying external actions.

Reference: https://learn.chatgpt.com/docs/app-server

No UI expansion or provider switch is authorised by this issue.

Do not implement this issue now: it is explicitly deferred at the owner's request. If work is reopened, start by reviewing the linked ChatGPT App Server documentation and resolve the hosted multi-user support, production status, authentication, isolation, credential storage, and recovery questions. Done means an approved implementation scope exists; this issue currently authorizes no UI or provider change.