250 models live right now

Every AI model on one key. Below list price.

Frontier models, image and video, and thousands of paid APIs for your agent. Pay per call, no subscription.

Get your key

Top up from $5 · no plan · a failed call is not charged

Why it costs less

Unused capacity, resold.

AI capacity is bought in bulk, and part of it sits idle. Mana resells that unused capacity to you, below list price. Same models, exact versions, nothing swapped.

Claude Fable 5.2−50%

$20.00$40.00

GPT-6 Astra−50%

$25.00$50.00

Grok 4.7−50%

$3.00$6.00

Mana List price · per 1M output tokens

Who it is for

Anyone who runs an agent.

  1. 01
    You build with AI

    Change three values: base URL, key, model ID. Your code stays.

  2. 02
    Your agent needs tools

    Find a lead, verify an email, search the web. Paid APIs per call, no monthly plan.

  3. 03
    You make images and video

    Image, video, voice and music models, on the same key and balance.

Works with the agent you already use

  • ChatGPT
  • Claude
  • Claude Code
  • Codex
  • Hermes
  • OpenClaw
  • Grok Bot
  • OpenCode
  • Cursor
  • Cline
  • Aside
  • Kilo Code
  • Continue
  • Open WebUI
  • LangChain

Start

Running in three steps.

  1. Get your key

    Sign up with your email.

  2. Top up from $5

    No subscription. Every call draws from your balance.

  3. Paste three values

    Base URL, key, model ID. Or let your agent set itself up.

Supercharge your agent.

One key, one balance, every model.

Get your key

Fuel for your agents.

The latest AI models and thousands of paid APIs on one key. Pay per call, no subscriptions: your agent gets more done from the first call.

Get your key

Every model, live

Charging the galaxy…

By provider

Ask any model anything.

Supply

Everything your agent needs. On one key.

01

Give your AI every frontier model.

Text, image, video, audio and decision models from every lab, on one key.

See catalogue
03

Run a thousand agents without running a server.

Each agent gets its own container and keeps its files. No hourly bill.

one container per agent 1 → 100 → 1,000 agents 1/4 vCPU · 1 GiB each
04

Give your agent powers a model does not have.

Verify an email, find a lead, post on X: 4,353 paid APIs, per call, with no subscription.

one cell per API email · leads · search · social per call · no subscription

One key

Five accounts, or one key.

You needWithout ManaWith Mana
Frontier modelsA model gatewayAt list priceUp to −58%Opus 5.5: $20 → $10 per 1M output
APIs, tools & MCPOne subscription per APIMonthly, used or not$0 / monthPay per call, nothing to subscribe
Images & videoAnother accountAnother invoice0% markupSame key, same balance
Agent runtimeAnother accountBilled by the hourBuilt inModel price +5%, no hourly bill
RoutingA router serviceOr rules you write and maintainBuilt inOne model name, the right model for each request
BillingFive accountsFive invoices1 balanceOne top-up, one invoice

Plug in

Made for your agent.

Your agent can sign up, get its key and ask you to pay, all by API. Or connect the one you already use: three values, and it runs on Mana.

  • ChatGPT
  • Claude
  • Claude Code
  • Codex
  • Hermes
  • OpenClaw
  • Grok Bot
  • OpenCode
  • Cursor
  • Cline
  • Aside
  • Kilo Code
  • Continue
  • Open WebUI
  • LangChain

Paste them into your agent.

Or by hand, three valuesBase URLhttps://mana.supply/v1+Your Mana key+A model IDclaude-opus-5.5

Or use Mana inside your assistant. Three apps for ChatGPT, Claude and Grok. Sign in with Mana, set a monthly limit, done.

All install steps →

Reviews

What builders say.

Pricing

Pay for what your agent uses.

Pay as you go

Top up from $5. Every call draws from it.

No subscription, no plan. Top up again when you need to.

  • Each call costs its live price, shown on every model
  • Set a monthly spending cap on each key
plannone
top upfrom $5
Get your key ↑
each calllive price
under list−58%

Questions

Questions

Do I have to change my code?

Three values: the base URL, your Mana key and a model ID. Requests use the OpenAI Chat Completions format, and Anthropic's /v1/messages works too.

Is it really the provider's model?

Yes. The provider's own model serves your request. No copy we host, no smaller model swapped in. Each listing names the exact version.

What does my agent's computer include?

Its own container with 1/4 vCPU, 1 GiB of memory and a disk. Claude Code, Codex or Grok CLI comes installed. The disk and the session last 7 days, so the next request picks up where the last one stopped. You pay the model's token price plus 5%, with no hourly charge.

Will thousands of tools flood my agent's prompt?

No. The MCP server gives your agent three tools, search, read and call, not one per API. Searching and reading are free.

What if a model or provider goes down?

Many models have more than one provider. If one fails, the next one answers, and you are not charged for the failed call.

Can you see my prompts?

They pass through our gateway on the way to the model, so they cross our servers while the request runs.

Supercharge your agent.

One key, one balance, every model.

Get your key
Get your key