How we compared them
This covers budget features only, not routing, logging or overall quality, and it comes from each product's public documentation on the date at the top of this page. We haven't tested these products under load. Features and plans change, so confirm anything you plan to rely on. Spend isn't in the table, because this page is about what these tools do. Spend is in early access: you can set budgets and see what each call cost today, and budgets that stop calls are being switched on across the platform.
At a glance
| LiteLLM | Portkey | Helicone | OpenRouter | |
|---|---|---|---|---|
| How you run it | Self-hosted proxy; budgets need a Postgres database | Managed gateway | Gateway, with limits set by a request header | Managed API |
| Where limits apply | Proxy, team, team member, user, key, customer; per-model on Enterprise | Provider integration and workspace | Global, per user, or per custom property | Each API key |
| Limit in | US dollars | US dollars (at least $1) or tokens (at least 100) | Cents or requests | Credits |
| Reset | Any window set by budget_duration; never, if unset | None, weekly or monthly | Rolling window of 60 seconds or more | Daily, weekly or monthly reset options |
| At the limit | Request rejected with a budget error | A cost limit expires the key | 429 | Not stated on the page we read |
| Availability | Open source; per-model budgets on Enterprise | Enterprise, and select Pro customers | Documented as a gateway feature | Set on key creation or update |
LiteLLM
The broadest set of levels, and you run it yourself. Budgets need a database: the docs say Budgets require a database
, and without one they aren't enforced. Keys have no budget by default, and without budget_duration a budget never resets.
Two details to note. If a key belongs to a team, only team and team-member budgets apply, so the key owner's personal budget isn't enforced. And budget reservation is on by default, which holds the estimated cost before a request goes to the provider and helps with parallel calls.
Portkey
Limits attach when you create a provider integration, and can also be set per workspace. You can have a cost limit, a token limit, or both, plus email alerts at thresholds you choose. Limits can't be edited: to change one, you duplicate the provider and set a new limit. Requests for models without Portkey pricing show as 0 cents and don't count toward the limit, which is worth checking if you use unusual models. Budget limits are on the Enterprise plan and for select Pro customers.
Helicone
Limits are expressed per request in a header, in the form [quota];w=[seconds];u=[unit];s=[segment]. With u=cents it limits spending, so 500;w=3600;u=cents;s=user allows $5 per user per hour. That's very flexible, and it means the caller sets the policy, so the code setting the header has to be code you trust rather than the agent itself. Over the limit, the request gets a 429.
OpenRouter
Each API key can have a credit limit, set at creation and changed later, and each key reports what remains. The documentation we read shows a daily reset at midnight UTC and mentions weekly and monthly resets. It's a simple way to get one limit per agent, by giving each agent its own key. The page we read suggests disabling keys that exceed limits yourself, so test what happens at the limit before relying on it.
How to choose
- You want to self-host and control everything: start with LiteLLM, and plan for running its database.
- You want one simple limit per agent across many models: per-key credit limits like OpenRouter's may be enough.
- You already log through a gateway and want limits by user or property: look at Helicone's policies.
- You're on an enterprise plan with workspaces: Portkey's limits fit that structure.
- You use one provider and have few agents: start with that provider's own limits, one project or workspace per agent.
Whichever you pick, test two things before trusting it: what happens at the limit, and what happens when the gateway can't check. See spend alerts vs hard limits for why that second test matters.
Questions
- Which LLM gateway has the best budget controls?
- It depends on your setup. LiteLLM has the widest set of budget levels if you can self-host. OpenRouter offers simple per-key limits. Helicone sets limits by header, by user or property. Portkey's budget limits are on Enterprise and select Pro plans.
- Does LiteLLM support per-key budgets?
- Yes. A virtual key can have a max_budget and a budget_duration. Keys have no budget by default, budgets need a Postgres database, and if a key belongs to a team only team and team-member budgets apply.
- Can Helicone limit spending?
- Yes. Its rate limit policy header can count cents rather than requests, for example 500;w=3600;u=cents;s=user for $5 per user per hour. Requests over the limit get a 429.