feat: add provider usage center

[lambertjosh]

What changed

Add a secret-free Provider Usage Center shared by the CLI, TUI, SDK, and VS Code profile. Core owns provider detection, normalization, source-scoped caching, in-flight coalescing, stale fallback, and failure isolation.

The provider-plan sources cover Kilo-managed MiniMax subscriptions through Cloud tRPC and direct MiniMax Token Plan credentials configured locally. Kilo Pass remains on the existing profile transport and is presented first in Plans & usage with its paid and free-bonus credits. Kilo-managed plans are labeled KILO GATEWAY; locally configured provider credentials are labeled DIRECT.

Managed MiniMax supports Plus, Max, and Ultra, preserves the unified text/image quota returned by MiniMax, and omits unsupported video-only quota rows. A managed card is detected only when the subscription has an enabled, read-only coding_plan BYOK entry; absent, disabled, user-managed, or unavailable routing metadata produces no managed card or quota request. Personal top-up state appears directly below Balance rather than among provider plans.

Architecture

flowchart LR
  subgraph Sources[Usage sources]
    Config["Local config and auth<br/>MiniMax Global / China credentials"]
    Cloud["Kilo Cloud tRPC<br/>plans, BYOK routing, managed quota"]
    Pass["Existing profile transport<br/>Kilo Pass, balance, top-ups"]
  end

  subgraph Gateway[packages/kilo-gateway]
    Trpc["Cloud tRPC client<br/>safe envelope parsing and projection"]
    Format["Shared window formatting<br/>localized label inputs"]
  end

  subgraph Core[packages/core]
    Detect["provider-usage.ts<br/>source detection and credential resolution"]
    Direct["minimax/usage.ts<br/>independent regional plan queries"]
    Managed["cloud.ts<br/>managed quota normalization"]
    Cache["Location-scoped ProviderUsage service<br/>TTL cache, coalescing, stale fallback,
    failure isolation"]
  end

  subgraph Contract[Public contract]
    Schema["packages/schema<br/>secret-free ProviderUsage schema"]
    Http["packages/opencode HTTP API<br/>GET usage / POST refresh"]
    SDK["packages/sdk<br/>generated typed client"]
  end

  subgraph Clients[User interfaces]
    TUI["CLI TUI Plans & usage<br/>load on open, Ctrl-R refresh"]
    VSCode["VS Code Profile<br/>load on open, Refresh button"]
  end

  Config --> Detect --> Direct --> Cache
  Cloud --> Trpc --> Detect --> Managed --> Cache
  Cache --> Schema --> Http --> SDK
  SDK --> TUI
  SDK --> VSCode
  Format --> TUI
  Format --> VSCode
  Pass -. "unchanged profile path" .-> VSCode

Loading

• packages/core is the source of truth for plan discovery, provider-specific normalization, and cache behavior.
• packages/schema is the client boundary: it contains normalized usage and structured periods, never credentials or provider-native payloads.
• packages/opencode exposes location-scoped read and refresh endpoints; packages/sdk generates the typed transport used by both clients.
• packages/kilo-vscode invalidates displayed usage on auth, organization, or workspace changes, then reloads only when the profile is open or the user refreshes.
• Direct MiniMax Global and China configurations remain independent plans and cache cells, even when they use the same credential.

Why

Users need one current-state view of provider plan quota without exposing credentials or coupling clients to provider-specific payloads. Billing history and subscription lifecycle remain in Cloud, organization-scoped Kilo data is skipped, and existing session/model accounting remains separate.

Dependency

Requires Kilo-Org/cloud#4220 for owner-scoped managed MiniMax quota.

[lambertjosh]

Minimax direct plan:

Note was not able to test subscription plan through Cloud, since my accounts don't have one.

[lambertjosh]

New version:

[lambertjosh]

When you don't have a kilo pass subscription:

[lambertjosh]

TUI: