Shared team wallet for LLM APIs: who actually has one, and what each cap misses

The question is usually asked as “what is the best LLM API aggregator with a shared team wallet?”, and the honest answer has a shape nobody’s pricing page admits: the wallet and the controls are mostly sold separately, and where they do meet, a gate is waiting. As of the docs on 27 August 2026: OpenRouter and Vercel AI Gateway have a genuine shared balance the whole team draws from. LiteLLM and Portkey have the deepest per-member and per-key budget machinery — and no wallet. OpenRouter is the one product that gets close to both halves on paper, and then caps the organization at ten members and puts workspace budgets and spend alerts behind Enterprise. So the real question is which half — or which gate — you can live with. That question, this piece can answer.

“Team wallet” is four different products

When the docs say team billing, they mean one of four things, and the differences cost real money:

If your team wants to stop juggling per-provider credit cards, only the first row solves that. If your problem is one intern burning $400 overnight, be careful with the first row — a shared balance is precisely the design where one person’s night can drain everyone’s month, and the member-level brakes vary sharply between the two products that offer it.

The two real wallets

OpenRouter Organizations is the purest shared wallet. The docs say it plainly: “All credits purchased in the organization account can be used by any organization member”, and API usage from any org key “is billed to the organization’s credit pool” (Organization management). Regular members can’t purchase credits or see billing — that stays with admins.

Now the fine print, priced. Topping up the pool by card costs 5.5% with a $0.80 minimum, crypto (USDC) 5% (FAQ) — the token prices themselves are passed through without markup, so the deposit fee is the price of the wallet. Unused credits can be expired after one year of purchase; that’s in the same FAQ. And organizations top out at ten members: “An organization can only have 10 members. Contact support if you need more.”

The control side is better than most people assume, which is why OpenRouter is the closest thing to a complete answer here. Keys take limit plus limit_reset — daily, weekly, monthly or none, resetting at midnight UTC (Management API) — and guardrail budgets attach a budget to a member, enforced per-user with the lower of key and guardrail limits winning. The gates sit one layer up: workspace budgets (daily/weekly/monthly/lifetime) are “available on the Enterprise plan”, and spend alerts are an email-only Enterprise private preview. We mapped the whole cap lattice in the per-key piece; everything there applies here unchanged.

Vercel AI Gateway runs the same prepaid model with a different trade. Credits sit on the team account, tokens carry “no markup and no platform fee” (pricing) — you pay the payment-processing fee yourself, or move to invoicing on Enterprise. The budget machinery is a proper three layers — team, project, API key, all enforced together: “A request has to pass every budget in scope, so if any one is exceeded, AI Gateway rejects the request even when the others have room” (Budgets). Cycles exist too: daily, weekly, monthly, or cumulative.

Two caveats, both from the same doc. First, budgets are declared soft: “the request that crosses the limit still completes and total spend can end up slightly over the budget.” Second, if you bring your own provider keys, that spend “isn’t counted in budgets” — BYOK traffic simply bypasses the brakes. And note what is still missing: the layers are team, project, key. Per member is not one of them.

The two real control panels

LiteLLM is the opposite animal: self-hosted proxy, your own provider keys, no money held anywhere — and the finest budget lattice of anything surveyed. Team budgets, member budgets inside a team (max_budget_in_team), per-key budgets, per-key TPM/RPM, and budget windows of any duration with documented reset times (team budgets, users). The precedence rule is worth quoting because it surprises people: “If a key belongs to a team, only the team (and team-member) budgets are enforced; the key owner’s personal budget does not apply.” The costs are operational, not financial: budgets require a database, you run the proxy, and model-specific budgets sit behind the Enterprise tag.

Portkey has the tidiest governance story — workspaces under an organization, budget limits in dollars or tokens, alert thresholds, resets weekly, monthly, or any custom 1 to 365 days via API (workspace budgets). It is also BYOK: Portkey bills you a subscription and your providers bill you for tokens, so there is no wallet to share. And the feature gate matters: both budget docs carry the same line — “Available on Enterprise plan and select Pro customers.” On the public pricing page, “Granular Budget & Rate Limits” sits in the Enterprise column; the $49/month self-serve tier (named “Production” there — the docs’ “Pro” and the pricing page’s tier names do not quite line up) does not list it.

Requesty sits between the camps and earns a paragraph: your org gets billed usage plus 5%, each key takes a monthly cap, and each user can get a monthly limit with hard cutoffs — “Spending cutoffs when limits are reached” (users). Per-user monthly limits with actual enforcement is the rarest feature in this survey. Groups and team budgets, though, are Enterprise-only, and 5% on every token is the steepest ongoing rate here — the OpenRouter deposit fee is paid once per top-up, this one compounds with usage.

The matrix

State-of-the-docs, 27 August 2026:

For contrast, the two big single-provider org tools do the same split internally: OpenAI Projects give org- and project-level hard limits that 429 on breach and reset monthly (spend limits); Anthropic Workspaces cap monthly spend per workspace, and per-user limits exist only in the special Claude Code workspace (Workspaces). Neither solves aggregation; both confirm the pattern — money pools and people-level brakes rarely ship together.

How to pick, by failure mode

Pick by the incident you are actually trying to prevent, and price the feature next to it:

• “Five people, one pile of credits, least ceremony” — OpenRouter Organizations. Budget the 5.5% deposit fee, set limit_reset on every key at creation time, add a guardrail budget per member, and remember ten members is the ceiling.
• “Shared pool, but I need budget cycles” — Vercel AI Gateway. Know that the caps are soft by design and BYOK traffic walks around them.
• “The wallet can stay with the providers; I need member-level enforcement” — LiteLLM if you will run infrastructure, Requesty if you would rather pay 5% than operate a proxy, Portkey if you are already an Enterprise conversation.

Full disclosure of the obvious: we sit in this market. The platform wallet we are building is designed to keep the prepaid pool and the per-key, cycle-based budgets in one control plane — that combination being missing is roughly why this article had to be a matrix instead of a recommendation. It is designed-as, not shipped; the current state is on our site and this paragraph is the only place we will mention it.

One habit transfers regardless of the pick: whatever enforces your caps, ask the two questions this survey kept reducing to — when does the check run (before the request, or after the spend lands — Vercel says plainly that the crossing request still completes), and what exactly is exempt (BYOK traffic, the default workspace, invoiced billing). Every product above has a clean answer to one of those questions written in its docs, and an uncomfortable one it leaves for you to find on the invoice. Read for the uncomfortable one first; it is the one that shows up as a number.