GPU Forge documentation
API base address: https://api.gpuforge.io. Every path in these docs starts with /v1/; send your API key as Authorization: Bearer <key>.
GPU Forge is a GPU cloud service for developers, small teams and AI agents. One account, one prepaid balance and one API give you two things:
- Ready AI programs. Open image and video models that we package, run and update: FLUX 3 Klein and Krea 2 (make and edit images), LTX-2.5 and MiniMax H3 (video with sound). You can send one request per result (a job) and pay only for the seconds it renders, or run a program on a machine of your own.
- GPU machines. Rent a GPU machine from Vast.ai, RunPod, Nebius or Jarvis Labs and run your own software on it, billed per second at a price you see before you start.
Everything works through a REST API, a command-line tool and a web console. AI agents can use GPU Forge with their own API key, inside the limits you set.
Core concepts
| Concept | What it means |
|---|---|
| Account | You sign in with Google or GitHub. Everything you run and pay for belongs to your account. |
| API key | A secret token (gf_...) that lets a script, agent or the CLI act for your account. Sent as Authorization: Bearer <key>. You create and revoke keys in the console. |
| Balance | Prepaid credit in US dollars. You add it by card (processed by Stripe). Usage is deducted as it happens. At $0 nothing new starts. |
| Program | Software that runs on a GPU machine. GPU Forge's ready programs are listed in the catalog. Your own programs are called capabilities: a description of what your software needs (GPU, disk, files) and how to start it. |
| Machine | A rented GPU computer running one program. The API calls a machine a pod and identifies it by pod_id. |
| Parked machine | A machine that was stopped with its disk kept. It costs only its storage price and wakes again in seconds instead of minutes. |
| Job | One request to a ready program that returns one result (an image or a video clip). GPU Forge finds or starts the machine for you; you pay for the render seconds only. |
Two ways to use the ready programs
| Jobs (pay per result) | Your own machine | |
|---|---|---|
| How | POST /v1/jobs with the program, route and input | Add the program to your account, start a machine, call the program's own API |
| You pay | The seconds the result took to render, at the program's hourly price (minimum $0.002 per job) | The machine's hourly price, per second while it runs, and its storage price while parked |
| Machines to manage | None | You start, park and release it (it also parks or stops itself when idle) |
| Best for | Occasional or bursty work | Steady, high-volume work |
Pages
| Page | What it covers |
|---|---|
| Quickstart | Sign in, add credit, create a key, run your first job, rent your first machine. |
| How GPU Forge behaves | Prepaid billing, prices before you run, how machines are chosen, parking, what happens at $0, limits for agents, security. |
| Authentication | API keys, sign-in, the CLI login, what a key may and may not do. |
| Jobs | The Jobs API: quote, create, wait for and list jobs. |
| Ready programs | The catalog and each program's routes, fields and responses. |
| Machines | Search offers, start, scale, watch, park, wake, release and destroy machines; SSH access. |
| Capabilities | Describe your own software once and run it on any provider; add a catalog program to your account. |
| Billing | Balance, top-ups, automatic top-up, daily caps, usage history, pricing, refunds. |
| Command-line tool | The gpuforge CLI for people and agents. |
| Errors | Error format, status codes and what to do. |
Machine-readable references
GET https://api.gpuforge.io/openapi.json: the OpenAPI description of the API.GET https://api.gpuforge.io/v1/agent-guide: a short plain-text guide written for AI agents (no key needed).GET https://api.gpuforge.io/v1/help: a full reference in one JSON response (needs a key).
Support
Email support@gpuforge.io. We aim to reply within one business day.
This page as Markdown, for tools and agents: /docs.md