A cross-platform desktop app that turns a trained model into a paid API endpoint. No payment gateway integration, no business registration in compliant zones, no platform between you and your customers.
The platforms you already know solve real problems. ApiCharge solves a different one: ownership.
Pick a serving framework — vLLM, ComfyUI, TorchServe, HuggingFace Diffusers, TensorFlow Serving — pick a GPU, deploy. Your model runs on infrastructure you chose, on a domain you control.
Set per-call, per-second, per-byte, or credit-based tiers per service. Change pricing whenever you like — existing customers keep the QoS they paid for, cryptographically. No platform commission on your inference revenue.
Customers pay directly into a Stellar wallet you own in USDC, EURC, or XLM. Under MiCA in the EU and GENIUS in the US, this is infrastructure — not a payment processor. No merchant account, no gateway integration.
The desktop app handles the messy bits — provisioning, container deployment, SSH, certificates — so you stay in modeling mode.
Pull from HuggingFace or Civit.ai, or import local files. Metadata travels with the model.
Browse live Vast.ai spot offers — filter by VRAM, region, reliability, hourly price. Or bring your own VM.
Per-call, per-second, per-byte, credit-based — multiple tiers per service. Configure once, change anytime.
One click. The container boots, gets a TLS cert, registers with the marketplace. Customers can pay.
Import models from HuggingFace Hub · Civit.ai · local filesystem
Pre-baked templates — no Dockerfile required. Bring your own works too.
A trained model is only useful if customers can call it. ApiCharge maps the inputs that already exist in your ComfyUI workflow or HuggingFace Diffusers config straight to typed API parameters — pick what you want exposed, name it, and you're done. No request schema to hand-roll, no glue code to maintain.
Each exposed parameter can be priced independently — gate higher resolutions or longer prompts behind a premium tier, keep the basic call cheap.
Define multiple tiers per service. Each tier carries its own rate-limiting strategy, cryptographically enforced at the proxy. Customers buy access tokens that lock in the QoS they paid for — your future pricing changes can never degrade their experience.
Two services on one server, priced independently. Server cost $0.35/hr from Vast.ai shown alongside your tiers — total transparency, configured directly inside the app.
When traffic comes, capacity comes. When it stops, costs stop. Run a single instance for a side-project, or autoscale across dozens of GPUs — same dialog.
Hosting platforms and model hubs have solved real problems for a lot of developers — discovery, onboarding, a pre-built audience. None of that is going away, and we're not asking you to leave.
But if you'd rather own the endpoint, set the price, and have customers pay you directly — without onboarding to a payment processor, registering a business in your jurisdiction, or letting a curation team decide what you're allowed to ship — ApiCharge is built for exactly that.
It's a beta — rough edges included — because we want feedback from people who'll actually use this. Log issues, file feature requests, tell us what's missing.
