For developers
Private inference with a drop-in API.
OpenAI-compatible endpoints that run on your own GPUs first, free, and fall back to the network with earned credits when a job outgrows them.
curl https://api.virtualgiant.com/v1/chat/completions \
-H "Authorization: Bearer $VG_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "[open model]",
"messages": [
{"role": "user", "content": "Summarize this contract."}
]
}'Already using an OpenAI-compatible client? Change the base URL and key.
Built for the work behind the work
Most Ai spend isn't the chat window. It's the volume underneath: the jobs that can run overnight and still need to be private.
Embeddings and search
Index a company's documents on your own GPUs, without sending names anywhere.
Document extraction
Pull fields from invoices, forms and contracts in bulk.
Summaries and classification
Work through tickets, emails and logs in bulk.
Evaluations
Run test suites against model outputs overnight on hardware you already own.
Images
Generate and edit images on certified machines.
Everyday chat
Interactive chat on open models, streamed at reading speed or better.
Choose a privacy tier per request
| Tier | Where it runs | What leaves your machines | Good for |
|---|---|---|---|
| Local | Your own GPUs | Nothing | Everything that fits; the default |
| Private | Your machines only, never the network | Nothing | Client records, HR, contracts |
| Network | Your machines plus certified network machines | The work, never your identifying details | Big jobs, big models, images and video |
Batch spends fewer credits
Fallback work that can wait costs fewer credits than interactive work, because it can run on any machine that's free.
Keep your frontier model
Send the hardest reasoning to the frontier model you already use, and the volume to VirtualGiant.
Every call accounted for
Itemized local and fallback usage per key, credit limits, and a ledger you can check.
Build on the first network.
Tell us what you'd run. Early developers help shape the API.