# Hayonix > GPU cloud: OpenAI-compatible inference, GPU and CPU pods, serverless workers, managed fine-tuning and batch inference, paid from one prepaid balance. The full documentation as one text file: https://hayonix.ai/llms-full.txt ## Get started - [Hayonix documentation](https://hayonix.ai/docs): What you can run, how to reach it, and where to begin: inference, pods, serverless workers, fine-tuning and batches on one prepaid balance. - [Set up a key and your environment](https://hayonix.ai/docs/get-started/setup): Create a project API key, set the three environment variables every tool reads, and install the SDK or CLI you want to use. ## Quickstarts - [Your first inference call](https://hayonix.ai/docs/quickstarts/inference): Send a chat completion to an open model with curl, the SDKs or the official OpenAI clients, then stream the answer token by token. - [Your first pod](https://hayonix.ai/docs/quickstarts/pods): Deploy a CPU pod from a public image, wait until it runs, run a command inside it, and terminate it so billing stops. - [Your first serverless endpoint](https://hayonix.ai/docs/quickstarts/serverless): Write a Python handler with the worker runtime, build it into an image, create an endpoint that scales from zero, run a job and read its streamed output. - [Your first fine-tune](https://hayonix.ai/docs/quickstarts/fine-tuning): Upload a small chat dataset, train a LoRA adapter on a base model, follow the job as it runs, and download the trained adapter. - [Your first batch](https://hayonix.ai/docs/quickstarts/batches): Upload a file of inference requests, run them as a batch at batch prices, wait for it to finish and download the results. - [Your first dedicated endpoint](https://hayonix.ai/docs/quickstarts/dedicated-endpoints): Serve a catalog model on replicas of your own, wait until it is serving, chat with it through the OpenAI-compatible API, and delete it. - [Your first code sandbox](https://hayonix.ai/docs/quickstarts/sandboxes): Create an isolated, short-lived sandbox, run Python, JavaScript and shell code in it, stream a run’s output, move files in and out, and end it so billing stops. - [Evaluate outputs with a judge model](https://hayonix.ai/docs/quickstarts/evaluations): Upload a dataset, have a judge model label, grade or compare every row, read the aggregate results and the rows that did not pass, and download the results file. - [Serve HTTP from a serverless endpoint](https://hayonix.ai/docs/quickstarts/http-endpoints): Run an image that serves HTTP itself behind a stable URL, scale it with requests in flight down to zero, roll releases without failed requests, and keep idle workers paused so the next request skips the cold start. - [Run a multi-node job on a cluster](https://hayonix.ai/docs/quickstarts/clusters): Create identical members on one private network across hosts, all at once or not at all, reach the head over SSH, start processes on every member with the launcher, run torchrun or mpirun across them, and share files on a shared volume. - [Cache models on hosts](https://hayonix.ai/docs/quickstarts/model-cache): Pin a model from the model hub to a commit, mount it into serverless workers and pods from the host’s model cache, and load it with networking off, so a worker on a host that already holds the model starts without downloading it. ## Concepts - [Projects and API keys](https://hayonix.ai/docs/concepts/projects-and-keys): How organizations, projects and keys relate, what a key may be allowed to do, and when a request needs X-Project-Id. - [Billing and credit](https://hayonix.ai/docs/concepts/billing): Prepaid credit, how amounts are written (integer nano-USD), how holds reserve credit for work in flight, and what happens when credit runs out. - [How compute is billed](https://hayonix.ai/docs/concepts/compute-billing): Per-second billing from the node agent’s own record of what ran, what a pod costs running and stopped, and how storage is metered. - [Storage and volumes](https://hayonix.ai/docs/concepts/storage): Container disk, pod volumes and network volumes: what each is for, what survives a stop or a configuration change, and where volumes live. - [Limits and quotas](https://hayonix.ai/docs/concepts/limits): Organization quotas on resources, inference rate limits per model, the rate-limit headers every response carries, and what to do when you hit one. - [Errors and retries](https://hayonix.ai/docs/concepts/errors): The one error envelope, its types and codes, the SDK error classes, and which failures are retried automatically. - [Idempotency](https://hayonix.ai/docs/concepts/idempotency): Make creating requests safe to retry with the Idempotency-Key header: what is stored, for how long, and the two ways a key is refused. - [Pagination](https://hayonix.ai/docs/concepts/pagination): Every list is cursor-paginated: read data and follow next_cursor until it is null, or let the SDKs iterate for you. - [Platform events](https://hayonix.ai/docs/concepts/platform-events): Webhook destinations that receive signed events when pods, jobs, models, batches, endpoints and billing change; verification, retries, pausing and secret rotation. - [Webhooks and signatures](https://hayonix.ai/docs/concepts/webhooks): Receive a signed HTTP callback when a serverless job finishes, verify its signature, and handle retries and duplicates. ## SDKs and CLI - [TypeScript SDK](https://hayonix.ai/docs/sdks/typescript): Install and configure the hayonix package, handle errors, tune retries and timeouts, paginate, stream, upload and download files, wait for resources and work with money. - [Python SDK](https://hayonix.ai/docs/sdks/python): Install and configure the hayonix package (sync and async), handle errors, tune retries and timeouts, paginate, stream, upload and download files, wait for resources and work with money. - [Use the SDK you already have](https://hayonix.ai/docs/sdks/compatible-sdks): Point the OpenAI SDK (chat completions or the Responses API) or the Anthropic SDK (the Messages API) at the inference API by changing only the base URL and the key: what each shape supports and what it refuses. - [Command-line tool](https://hayonix.ai/docs/sdks/cli): Install the hayonix command, sign in, script it with --json and -q, set up shell completion, and read its exit codes. ## API reference - [API reference](https://hayonix.ai/docs/reference): every control-plane operation, grouped by area - [Inference API (OpenAI-compatible)](https://hayonix.ai/docs/reference/inference): The OpenAI-compatible routes: chat completions, completions, embeddings, rerank and models, with authentication, billing, limits and errors. - [Speech, transcription and images (OpenAI-compatible)](https://hayonix.ai/docs/reference/audio-and-images): Text-to-speech, speech-to-text, translation and image generation on the OpenAI-compatible routes, with how each is measured, billed and limited. - [Account](https://hayonix.ai/docs/reference/account): 2 operations: the signed-in user, memberships, projects and pending invitations; the credential making the request, with its org, project and permissions - [Organizations](https://hayonix.ai/docs/reference/organizations): 4 operations: list the organizations you belong to; create an organization (you become its owner; a default project is created); get an organization; update an organization - [Members](https://hayonix.ai/docs/reference/members): 4 operations: list members; get a member; change a member's role; remove a member - [Invitations](https://hayonix.ai/docs/reference/invitations): 4 operations: list invitations; invite someone by email (a pending invitation to the same address is replaced); cancel a pending invitation; accept an invitation addressed to your verified email - [Projects](https://hayonix.ai/docs/reference/projects): 6 operations: list projects; create a project; get a project; update a project; archive a project (projects are never deleted); restore an archived project - [API keys](https://hayonix.ai/docs/reference/api-keys): 5 operations: list API keys in the project; create an API key (the secret is returned once); get an API key; rename an API key; revoke an API key (takes effect immediately; idempotent) - [SSH keys](https://hayonix.ai/docs/reference/ssh-keys): 5 operations: list SSH keys; add an SSH public key; get an SSH key; rename an SSH key; delete an SSH key - [Secrets](https://hayonix.ai/docs/reference/secrets): 5 operations: list secrets; create a secret; get a secret (metadata only); rename a secret or replace its value; delete a secret - [Registry credentials](https://hayonix.ai/docs/reference/registry-credentials): 5 operations: list registry credentials; add registry credentials; get registry credentials (without the password); update registry credentials; delete registry credentials - [Audit logs](https://hayonix.ai/docs/reference/audit-logs): 3 operations: list audit log entries; export audit log entries as CSV; get an audit log entry - [Billing](https://hayonix.ai/docs/reference/billing): 12 operations: get the credit balance; billing settings, credit standing, spend limits, auto-recharge and limits; set the low-balance alert threshold and spend limits (null clears); the rate-limit level: per-minute limits in force and what reaches the next level; turn auto-recharge on or off (threshold → target, monthly cap, saved card); start a top-up: returns a payment with the checkout URL to send the customer to; … - [Pricing](https://hayonix.ai/docs/reference/pricing): 1 operation: current public prices (no authentication) - [Pods](https://hayonix.ai/docs/reference/pods): 18 operations: hardware for pods: per SKU, its prices and what can be launched now in each region; price estimate for a pod configuration: hourly running and stopped cost; list pods in the project (terminated pods are left out unless asked for); create a pod from a template or an image (launch gate and quotas apply); get a pod: its state, configuration and how to connect; rename a pod (does not touch the container); … - [Templates](https://hayonix.ai/docs/reference/templates): 5 operations: templates you can launch: official, public and your organization’s private ones; create a private (organization) or public template; get a template; update one of your organization’s templates (pods made from it are not changed); delete one of your organization’s templates (pods made from it keep running) - [Volumes](https://hayonix.ai/docs/reference/volumes): 8 operations: list live volumes in the project; create a network volume (billed per GB-second until deleted); get a volume, where it is pinned and which pods mount it; rename a volume; delete an unattached volume and its data (irreversible; billing stops now); grow a volume (sizes never shrink); … - [Serverless](https://hayonix.ai/docs/reference/serverless): 21 operations: list serverless endpoints in the project; create a serverless endpoint (the webhook secret is returned once); get a serverless endpoint; update an endpoint. Changing the image, command, environment, SKUs, GPU count, concurrency, disk or HTTP settings creates a new release; workers are replaced progressively (busy ones finish first; an HTTP endpoint keeps old workers serving until new ones are healthy).; delete an endpoint: its jobs are canceled and its workers removed; replace the webhook signing secret (returned once); … - [Models](https://hayonix.ai/docs/reference/models): 3 operations: list catalog models (public ones; with credentials, also the project's private ones); get a catalog model (private models only for their own project); mint a short-lived playground token for the inference API (signed-in users only) - [Model references](https://hayonix.ai/docs/reference/model-references): 4 operations: list the project’s model references; accept a model reference (checked against the model hub, pinned to a commit); get a model reference and the files it resolved to; delete a model reference nothing mounts - [Dedicated endpoints](https://hayonix.ai/docs/reference/dedicated-endpoints): 17 operations: list dedicated endpoints in the project, newest first; create a dedicated endpoint (a catalog model or your own Hugging Face model) with its first deployment; get a dedicated endpoint with its deployments and replica counts; update the display name or the inactivity timeout; delete an endpoint: its replicas are removed and its model ID stops resolving; start a stopped endpoint (needs credit; replicas start and are billed once ready); … - [Files](https://hayonix.ai/docs/reference/files): 8 operations: list files in the project; upload a file (multipart form: file, purpose); validated asynchronously; the upload size cap, the purposes and the validation limits; get a file, with its validation state and report; delete a file; download a file's content; … - [Fine-tuning](https://hayonix.ai/docs/reference/fine-tuning): 16 operations: models that can be fine-tuned, how, on which hardware and at what price; price estimate for fine-tuning on a validated dataset; list fine-tuning jobs, newest first; create a fine-tuning job (reserves its estimated cost; 402 when credit is short); get a fine-tuning job; cancel a job (charged only for the tokens it trained; idempotent); … - [Batches](https://hayonix.ai/docs/reference/batches): 5 operations: list batches in the project, newest first; create a batch from an uploaded file of requests (validated and run asynchronously); get a batch: status, request counts, stage times, cost so far; list the lines of a batch in input order: status, error, tokens and cost; cancel a batch (requests in flight may finish; completed ones are kept and charged) - [Evaluations](https://hayonix.ai/docs/reference/evaluations): 7 operations: list the models this project can use as a judge or to generate candidates; estimate an evaluation: the most it can cost, and its first row’s requests; list evaluations in the project, newest first; create an evaluation over an uploaded dataset (validated and run asynchronously); get an evaluation: status, progress, results so far, cost; cancel an evaluation (calls in flight may finish; finished rows are kept); … - [Sandboxes](https://hayonix.ai/docs/reference/sandboxes): 15 operations: sandbox sizes with their prices, and the limits that apply to sandboxes and runs; list sandboxes in the project, newest first (ended ones only when asked for); create a sandbox (launch gate and quotas apply); it ends by itself at its deadline; get a sandbox: its state, time left and cost so far; move its deadline later (never past its maximum lifetime); end a sandbox now: runs are cut off, its files are deleted, billing stops; … - [Clusters](https://hayonix.ai/docs/reference/clusters): 8 operations: what a cluster of this shape costs per hour at today’s prices; list clusters in the project, newest first (deleted ones only when asked for); create a cluster: every member is placed on a host at once, or nothing is created; get a cluster: its state, its members (rank, host, address, state) and how to connect; rename a cluster (nothing else about a cluster changes after creation); delete a cluster: every member, its network and its shared volume are destroyed; … - [Usage](https://hayonix.ai/docs/reference/usage): 3 operations: spend over time, grouped, with a breakdown table; the cost report as CSV; the current month's spend by product - [Webhooks](https://hayonix.ai/docs/reference/webhooks): 11 operations: every event type a webhook can receive, with the JSON Schema of its data; list the project’s webhook destinations, with their delivery health; create a webhook destination (its signing secret is returned once); get a webhook destination; change a webhook destination, or pause or resume it; delete a webhook destination and its delivery log; …