MODEL CATALOGUE
A gateway designed to separate model access from payments and accounting. The current implementation contains one internal offline Mock adapter; external model integrations remain planned.
The names below describe integration targets. Public availability and verified tariffs are shown separately from development status.
SOVREG
Deterministic text extraction
An offline, deterministic extractive text adapter used for internal testing. It counts whitespace-separated words as simulated tokens. It is not a language model and has no public generation endpoint.
OpenAI
Text & code · intended integration
Planned gateway support for the GPT family. The operator has not selected a public model catalogue, confirmed tariffs or enabled commercial access.
Anthropic
Text & code · intended integration
Planned gateway support for the Claude family. Exact model IDs, availability and provider terms will be reviewed before integration.
Task-specific AI · intended integration
Planned gateway support for the Gemini family. Features and pricing will be published only after a specific integration is verified.
DeepSeek
Text & code · intended integration
Planned gateway support for DeepSeek models. Availability and supported capabilities will depend on the model and API approved for use.
API
Additional adapters · planned
The gateway architecture is intended to accept additional provider adapters. Compatibility must be implemented and tested for each integration.
Planned integrations do not imply provider partnerships or access to every model. No external AI provider is enabled. Mock word counts must not be used as real LLM token measurements.
Future model selection can consider the task, price, latency and enabled capabilities. A model becomes available only after its adapter, usage reporting, provider access and financial controls have been tested.