Is the DreamPrompting LLM API really free?

Yes. There is no card, no subscription and no per-token charge. It is free because the gateway owns no models: it forwards every request to an upstream provider's own free tier, so what you are using is a shared allowance rather than a purchased one.

What are the rate limits?

100 requests per minute per IP address, plus a rolling 24 hour quota of 5,000 requests and 500,000 tokens per account.

Is the quota per key or per account?

Per account. It is summed across every key you hold, active or revoked. Counting per key made the budget resettable by minting a new one.

What happens when I hit a limit?

You get an HTTP 429 whose message names which of the two budgets ran out. The window is rolling rather than daily, so capacity returns gradually instead of at midnight.

When should I not use this API?

When you need guaranteed capacity, an uptime SLA, data residency, or traffic that must never be rate limited. For those, bring your own upstream provider key or self-host the router.

Pricing

Name: DreamPrompting Free LLM API
Brand: DreamPrompting
Availability: InStock

The free plan costs nothing, and this page is its small print. Pro removes the daily quota for $9 a month.

per request, per key, per month.

Get a free key

upstream providers behind one key

live and serving right now

card details we store

Why it is free, which is also why it has limits

This gateway owns no models and rents no GPUs. Every request is forwarded to an upstream provider's own free tier, and when that provider is busy or exhausted the router fails over to the next one. That is the entire business model, and it is also the entire limitation: you are sharing an allowance, not buying one.

The limits below exist so that one caller cannot spend the allowance everyone else is using. They are not a paywall with a number on it. The free plan does not get smaller to make Pro look better.

The limits

Burst

100/min

Per IP address. Stops one machine from crowding the endpoint.

Requests

5,000/day

Per account, on a rolling 24 hour window. Requests served by your own provider key are not counted at all.

Tokens

500,000/day

Per account, rolling. Prompt and completion together. This is usually the one you meet first.

Input

32,000

Maximum tokens in one request, counted before routing.

Output

8,192

Ceiling on max_tokens, applied whether or not you set one. A single unbounded request can drain a provider's whole daily allowance.

Stored context

12,000

Token budget for a server-side conversation, if you use context ids instead of resending history.

Every number on this page is rendered from the constant that enforces it, so it cannot drift out of date. If you need more, add your own provider key, which is free, or get Pro.

Pro: no daily quota, $9 a month

No daily request or token quota.
Fair use: 30,000,000 tokens in a rolling 30 days. That is 60 times the free daily budget. Write to us if you need more.
The same 27 providers and the same models as the free plan.
Prepaid for 30 days. Nothing renews by itself, so there is nothing to cancel.

for 30 days

What Pro does not buy. We do not own the models. Pro removes our daily quota, which we control. It does not make an upstream provider answer when every provider is busy or down. We do not refund a bad hour. If the service is bad for a week, we add a free month to your account.

The free way to get the same thing. Add your own provider key. Requests that go through it do not count against any quota, and it costs nothing. Pro is for people who do not want to manage provider accounts.

Payment is by card or M-Pesa through Paystack. Our Paystack account settles in Kenyan shillings, so the charge is KES 1,170. Your bank converts it to your currency. Card details go to Paystack and never reach our server.

The quota is per account, not per key

Usage is summed across every key on your account, including ones you have revoked. That is deliberate, and it is a fix rather than a design: counting per key made the daily budget resettable in about ten seconds. Mint a key, spend its tokens, revoke it, mint another. Keys are credentials, not allowances, and someone holding ten of them has one budget rather than ten.

You can see where you stand at any time, without reading a dashboard:

curl https://dreamprompting.com/api/v1/quota \
  -H "Authorization: Bearer YOUR_API_KEY"

Cross either budget and you get an HTTP 429 whose message names which one ran out, because "rate limited" on its own tells you nothing you can act on. The window is rolling rather than calendar-daily, so capacity comes back gradually through the day instead of all at once at midnight.

When you should not use this

Most free-tier pages stop before this section. Here is ours.

Traffic that must never be rate limited. Free tiers run out, and when the upstreams are exhausted the honest answer is a 429 rather than a queue.
Anything needing an uptime guarantee. There is no SLA, and we are not going to invent one. What there is instead is failover across 27 providers and a public record of what has actually gone wrong.
Data residency or compliance requirements. Your request goes to whichever upstream answers, in whichever country that upstream runs.
Work where a substituted model breaks you. Automatic routing can answer from a different model than the last request did. Send "allow_fallback": false to pin one, and accept an error instead of a substitute.

What to do instead: bring your own key

Most of the providers here run a permanent free tier of their own. Get a key from one, add it to your account, and your requests go through it instead of the shared pool. That means no daily quota at all, because the quota rations our allowance and you are no longer using it. It is free, it takes about a minute, and you can remove the key whenever you like.

It is also the neighbourly thing to do: every request running on your own key is one that is not competing with everybody else for the same free tier.

The routing layer itself is open source, so you can run this exact failover logic against your own keys: openai-shim, MIT, no dependencies. The write-up explains what it does and why.

Your data

Worth knowing before you send anything sensitive:

Requests are forwarded to third-party providers, whose own terms then apply to that copy of your prompt. The models page names every provider a request can reach.
How we handle your data is set out in the privacy policy.

If a third party holding a copy of your prompt is not acceptable, the honest advice is the same as the section above: run it yourself.

Questions

Is there a paid tier coming?

Not today, and there is no waiting list to join. If one ever exists, everything described on this page stays free.

Why do I need an account at all?

Because the quota has to attach to something. You can try the API with no account at all in the playground, which is IP limited instead.

Can I use my own provider API key?

Yes, and it removes the daily quota entirely. Add one to your account: it is checked against the provider once, encrypted, never shown again, and only ever used for your own requests. Remove it and you go straight back to the shared pool.

Can I raise my limits?

Ask. Higher limits are set per account and cost nothing. Tell us what you are building and roughly what volume you expect.

Which models do I get?

Whatever the upstreams are serving, which changes. The models page shows what is routable right now and which providers are live.

Do you charge for streaming or tool calls?

No. Nothing here is metered by feature. Streamed requests count their real tokens against your quota exactly like any other request.

Start

Get a free key, then point any OpenAI client at https://dreamprompting.com/api/v1. The API docs have the rest.