Get tokens. Connect to our Local LLM in minutes.
Register, pick a token package and call an OpenAI-compatible endpoint running on our own AMD ROCm GPU server — private models, transparent metering and no third-party API in the loop.
Last updated: 2026-07-16. Provider: SnapSiteBuild AI Centre of Excellence — founder Krish Chimakurthy.
TL;DR — The SnapSiteBuild Developer API is a hosted, OpenAI-compatible local LLM endpoint that runs on our own AMD ROCm GPU server (node amd-rocm-01 ), not a third-party API behind a reseller. You register, choose a token package, generate an API key, and call the standard /v1/chat/completions endpoint with any OpenAI SDK or tool — Cursor, Claude Code, LangChain, LlamaIndex. You get open models, transparent per-token metering, private inference, and a representative cost up to roughly 96% lower than the large clouds. The Sandbox tier is free, no card required.
What the SnapSiteBuild Developer API is
It is a self-hosted LLM API exposed as a private LLM endpoint at https://api.snapsitebuild.com/llm/v1 . "OpenAI-compatible" means the request and response shapes match the OpenAI Chat Completions interface: a Bearer token in the Authorization header, a JSON body with model and messages , and a response that includes choices and a usage object with prompt, completion and total token counts. If your code already talks to OpenAI, it talks to us after a one-line base-URL change.
The difference is where it runs. Every token is generated on GPUs we own and operate, so there is no per-seat enterprise contract, no opaque markup, and no customer data flowing to an external model vendor. It is the same hosted open model API convenience as a cloud — with the economics and privacy of running the silicon yourself. For teams that later want this inside their own four walls, the same stack underpins our private local LLM setup and enterprise AI gateway .
Connect in three steps
Register with a work email. No card is needed for the free Sandbox package.Pick a token package — choose your token volume, requests-per-minute rate limit, context window and model access.Generate an API key and call the OpenAI-compatible endpoint on node amd-rocm-01.
A minimal request looks exactly like an OpenAI call, just pointed at our base URL:
curl https://api.snapsitebuild.com/llm/v1/chat/completions \
-H "Authorization: Bearer ssb-live-YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "llama3.1:8b",
"messages": [{"role": "user", "content": "Hello from SnapSiteBuild"}]
}'The response carries a usage block — for example "prompt_tokens": 12, "completion_tokens": 38, "total_tokens": 50 — which is exactly what we meter against your package, so billing is never a black box. To use an existing SDK, override the base URL and key instead of OpenAI's: in the OpenAI Python or Node client set base_url to https://api.snapsitebuild.com/llm/v1 and api_key to your ssb-live- key. Nothing else in your code changes.
Cheaper than OpenAI, Anthropic and Google
Because inference runs on our own AMD GPUs, the per-token cost for comparable open models is a fraction of the big clouds. The table below uses representative public list prices, in USD per 1M tokens, for illustration.
Provider · Model Input / 1M Output / 1M You save
SnapSiteBuild Local LLM · Qwen3-Coder-30B (self-hosted)$0.15 $0.45 Baseline OpenAI · GPT-4o $2.50 $10.00 ~95% less Anthropic · Claude Sonnet 4 $3.00 $15.00 ~97% less Google · Gemini 2.5 Pro $1.25 $5.00 ~90% less
Representative/illustrative public list prices, USD per 1M tokens, as of June 2026. Cloud rates vary by tier, region and context length; figures are shown for comparison only. Savings depend on workload, but the per-token gap commonly lands up to roughly 96% lower on our AMD GPU node.
Models you can call
The endpoint serves open models self-hosted on our AMD ROCm cluster. Available today:
llama3.1:8b — fast general chat and drafting, ideal for the free Sandbox and prototyping.
Rolling out on the same private endpoint (the free Sandbox stays free as these land):
Qwen3-Coder-30B Code generation, refactoring and tool-calling for coding agents.
DeepSeek V4-Flash Low-latency, high-throughput responses for chat and classification.
Nemotron 3 Nano / Super (NVIDIA) Agentic workloads with strong instruction following.
GLM-5.2 Reasoning-heavy tasks and longer multi-step prompts.
Need a model we don't list yet, or a fine-tune of your own? That is what a dedicated node is for — see the CTA below.
Token packages, metering and quotas
Packages are defined by four levers — monthly token volume, requests-per-minute rate limit, context window and which models you can reach — not by a confusing matrix of add-ons. The entry tier is Sandbox , which is free and needs no card, so you can test the endpoint and SDK integration before committing. Higher-volume packages raise your token allotment, rate limit and model access; you can switch packages at any time. Usage is metered per token: every call returns its exact prompt, completion and total token counts, and your package meter draws down against the total. If you outgrow shared capacity or need hard per-team quotas and routing, that logic lives in our enterprise AI gateway .
Is your data private?
Yes — the whole point of a self-hosted endpoint is that prompts and completions are processed on hardware we control, with no third-party model API in the loop. We do not sell your prompts, and we do not route them to OpenAI, Anthropic or Google to fulfil a request. For regulated workloads that cannot share infrastructure at all, the same models can be deployed inside your own environment via our local LLM setup , keeping every token on your network. If you are grounding answers in sensitive documents, pair the API with a RAG knowledge base so retrieval and generation both stay private.
Tools and SDKs that work
Anything that speaks the OpenAI Chat Completions protocol works: the official OpenAI Python and Node SDKs, Cursor (set a custom OpenAI base URL and key), Claude Code and other agents via an OpenAI-compatible base URL, plus LangChain, LlamaIndex, Vercel AI SDK, LiteLLM and Continue. That makes the endpoint a drop-in backend for AI agent development and copilots. Before you ship, validate model behaviour, tool-use correctness and prompt-injection resistance with structured LLM testing and evaluation — non-determinism is the part of AI delivery most teams under-test.
Developer FAQ
Is the API OpenAI-compatible?
Yes. The /v1/chat/completions interface matches OpenAI — Bearer-token auth, the same JSON request body and a response with choices and a usage object. Point any OpenAI client at our base URL and use your ssb-live key as the bearer token.
Which SDKs and tools work — does it work with Cursor and Claude Code?
Any OpenAI-compatible client: the OpenAI Python/Node SDKs, Cursor, Claude Code, LangChain, LlamaIndex, LiteLLM, Continue and the Vercel AI SDK. Set the base URL to https://api.snapsitebuild.com/llm/v1 and the API key to your ssb-live key.
How much cheaper is it than OpenAI, Anthropic and Google?
Because inference runs on our own AMD GPUs, per-token cost for comparable open models is typically up to roughly 96% lower than the large clouds — see the representative comparison table above. Actual savings depend on your model and workload.
Which models can I call?
llama3.1:8b is live today, with Qwen3-Coder-30B, DeepSeek V4-Flash, Nemotron 3 Nano/Super and GLM-5.2 rolling out on the same endpoint. All are open models self-hosted on our AMD ROCm cluster; custom models are available on a dedicated node.
How is metering and quota handled?
Per token. Each response returns its exact prompt, completion and total token counts, and your package meter draws down against the total. Packages set your monthly token volume, requests-per-minute rate limit, context window and model access, and you can switch tiers at any time.
Is my data private?
Yes. Prompts and completions are processed on GPUs we own, with no third-party model API in the loop and no selling of your prompts. For fully isolated deployments, the same models can run inside your own infrastructure.
Need a dedicated AMD node?
For reserved GPU capacity, custom or fine-tuned models, SSO, audit logs and an SLA on your own private endpoint, talk to us about a dedicated AMD node . We will size the hardware to your throughput and provision an isolated endpoint — the same OpenAI-compatible API, dedicated to you. Explore the wider platform across our AI services or start free on the Sandbox tier today.
Explore SnapSiteBuild AI services Contact SnapSiteBuild: Customersupport@snapsitebuild.com