Service Overview¶
What Nexus Is¶
Nexus accepts requests from your client, validates the key, sends the request to a compatible model, and returns the response in a familiar format.
For users, this means:
- no need to set up separate foreign access;
- you can keep using familiar CLI tools;
- the key, balance, and model list are managed in one place;
- support is available in Russian, and documentation is available in Russian and English.
How Access Works¶
Each Nexus API key is configured separately. Every key has:
- its own list of allowed endpoints;
- its own list of allowed models.
The balance is tracked at the Nexus profile level, and multiple keys on the same profile spend the shared profile balance.
That means two keys can see different models and different endpoints. Always check the current list with GET /v1/models.
What Is Supported¶
Nexus supports:
GET /v1/modelsto check which models your key can use;GET /v1/balanceto check the current profile balance;- Realtime-compatible WebSocket clients via
GET /v1/realtime; - Claude Code via
POST /v1/messages; - token counting for Claude-compatible clients via
POST /v1/messages/count_tokens; - Codex via
POST /v1/responses; - compact response workflows via
POST /v1/responses/compact; - OpenAI-compatible clients via
POST /v1/chat/completions; - legacy OpenAI-compatible workflows via
POST /v1/completions.
The detailed endpoint list is on the Supported Endpoints page.
Public status for the Nexus API, model compatibility, documentation and Telegram bot is available at status.nexus-hub.ru.
Streaming¶
Streaming is supported on:
POST /v1/chat/completions;POST /v1/messages;POST /v1/responses;POST /v1/responses/compact.
How Pricing Works¶
Pricing is based on actual model usage: input tokens, output tokens, and supported cache types. The Anthropic-compatible API supports explicit 5-minute, 1-hour, and mixed cache lifetimes; ambiguous cache usage is held for safe reconciliation.
Before the request starts, Nexus temporarily reserves an amount equal to the maximum possible request cost. After the request completes, the reserve is released and the actual cost in rubles is charged. If the cost cannot be calculated reliably, the request is sent for manual review.
GET /v1/realtime follows the same conservative approach: Nexus may create a
temporary reserve before the first billable response.create, charges
automatically only with reliable final usage, and sends ambiguous sessions for
manual review.
More details: Pricing and Balance