Multi-provider plan support: track Claude Max + ChatGPT Plus (Codex) overage in parallel

[samshennan]

Summary

codeburn plan set <id> currently accepts only one plan at a time, scoped to a single provider. Users who run multiple flat-rate subscriptions in parallel (e.g. Claude Max + ChatGPT Plus for Codex) cannot get accurate "API-equivalent vs subscription overage" attribution because only one of the two can be active.

Why it matters

The headline cost figure CodeBurn shows is tokens × LiteLLM rate. Without a registered plan, that number wildly overstates real cash outflow for anyone on subs. Today CodeBurn only solves this for one provider per machine.

My setup (representative)

• Claude Code Max — $200/mo flat (already supported via claude-max)
• ChatGPT Plus → Codex CLI / GPT-5.4-5.5 — $20/mo flat
• Some metered Gemini / Anthropic raw API on top

After codeburn plan set claude-max:

• Claude figures correctly switch to overage-aware view.
• Codex figures still show as raw API-pricing — $8.26 for GPT-5.4 in 7 days, $46.67 for GPT-5.5, etc. — when the real cost was the $20 flat sub.

Repro

codeburn plan set claude-max
codeburn plan set custom --monthly-usd 20 --provider codex   # there's no clean way to add a second plan
codeburn plan                                                 # still shows a single plan

Suggested shape

1. Make plans a list keyed by provider rather than a single scalar.
2. codeburn plan add <id> [--provider <p>] [--monthly-usd <n>] [--reset-day <d>] and codeburn plan remove <provider>.
3. Add a built-in chatgpt-plus plan id ($20/mo, provider: codex) so users don't have to roll a custom every time. This pairs naturally with claude-pro / claude-max / cursor-pro.
4. Optional: codeburn plan list showing the active set with current month overage per plan.

Bonus

Cursor Pro + Claude Max is the same shape — same fix unlocks that combo too.

Happy to test a patch.

[samshennan]

After reading the source, the constraint is structural:

• src/plans.ts:1 imports Plan, PlanId, PlanProvider from config.js as single-valued types.
• src/plan-usage.ts:11 passes a single plan: Plan through PlanUsage, and the storage in ~/.config/codeburn/config.json keeps one plan object.
• PRESET_PLANS (plans.ts:6-31) is keyed on the four singleton plan ids. There's no concept of "the active set of plans".

To support Claude Max + ChatGPT Plus in parallel, the refactor would need:

1. Storage: config.plan: Plan → config.plans: Plan[] (with a migration that wraps the existing scalar).
2. CLI: replace plan set <id> with plan add <id> / plan remove <provider> / plan list. Backwards compat: keep set as alias that replaces a same-provider plan.
3. Aggregation: in computePlanUsage, partition ProjectSummary[] by provider (already on ParsedApiCall) and run the existing overage math per plan.
4. New preset: chatgpt-plus ($20/mo, provider: codex). This is the most-requested combo (Claude Max + ChatGPT Plus + Codex CLI).

The display surface (status, report headers) needs to render N plans rather than 1, but the per-plan computation in plan-usage.ts is reusable as-is once you partition by provider.

Smallest user-visible wedge that doesn't require the full refactor: just add the chatgpt-plus preset and let users alternate which plan is active. Helps very little — the whole point is having both in view simultaneously — but it's a 5-line PR if you want a stopgap.

I'm on Claude Max + ChatGPT Plus + occasional metered Anthropic and would happily verify a patch.