ShvIA
TIMELINE
SIGN IN

Docs · Start here

ShvIA developer documentation

ShvIA is an AI gateway: one account, one API key and one usage record in front of models that run on the operator's own GPUs and on cloud providers. This documentation covers the HTTP API you use to call it from your own code and tools.

The hosted service lives at https://ai.shvia.org, and https://ia.shvia.org serves the same application. Every example here uses the first one.

Three ways in#

All three take the same API key, and their chat calls are metered together.

SurfaceEndpointsUse it when
OpenAI-compatible/v1/chat/completions, /v1/responses, /v1/models, /v1/audio/*your tool or SDK already speaks the OpenAI API
Anthropic-compatible/v1/messagesyour tool or SDK speaks the Anthropic Messages API, such as Claude Code
Native/api/v1/*you want what the ShvIA apps use: stored conversations, attachments, reasoning effort, usage and cost

What the gateway adds#

  • One key for many models. Models on the operator's GPUs and on cloud providers share one model list and one request format.
  • Your own provider keys. For cloud models you bring your own provider key (BYOK), so the contract and the bill stay between you and the provider. See providers and your own keys.
  • Per-answer accounting. Every chat call is metered in tokens and time, against limits set per account group. Embeddings and audio are not metered. See usage and cost and rate limits and quotas.
  • A stated data path. Each model declares whether a prompt stays on the operator's infrastructure or leaves it. See your data.

Start here#

OpenAI-compatible API#

  • Models: GET /v1/models, model ids, and how to find the models you can actually use.
  • Chat completions: POST /v1/chat/completions, with streaming and tool calls.
  • Responses: POST /v1/responses, the stateless slice.
  • Audio: speech to text and text to speech.

Anthropic-compatible API#

  • Messages: POST /v1/messages and token counting.

Native API#

  • Chat stream: POST /api/v1/chat over server-sent events, and the synchronous POST /api/v1/agent/chat.
  • Conversations: list, read, export and delete stored conversations.
  • Embeddings: POST /api/v1/embeddings.
  • Collections: send your own documents once, and name them in a chat call so the model answers from them.
  • Usage and cost: quota windows, spend by provider, and per-conversation cost.

Reference#

Protocols#

Guides#