loading
loading
Neko-AI is an OpenAI-compatible gateway. Requests route across health-checked upstream accounts, the tokens that actually stream are billed, and the unused reservation goes back to your balance.
sk-neko-… · credits per 1M tokens · reserve → settle
curl https://inference.neko-build.tech/v1/chat/completions \
-H "Authorization: Bearer $NEKO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4.1-flash",
"stream": true,
"messages": [{ "role": "user", "content": "ping" }]
}'data: {"choices":[{"delta":{"role":"assistant"}}]}
data: {"choices":[{"delta":{"content":"pong"},"finish_reason":"stop"}],"usage":{"prompt_tokens":9,"completion_tokens":1}}
data: [DONE]Four behaviours, all of them observable from your side of the API.
Chat completions, Responses and Anthropic Messages share one base URL, so an existing SDK only needs a new host and key.
Prices are credits per 1M tokens. Each request reserves its ceiling, settles on real usage, and refunds the difference.
GET /v1/models lists every enabled model with its context window and price. Public ids are stable across upstream changes.
The SSE stream is decoded once: the same frames feed your client and the billing parser, so a stream bills exactly what it delivered.
Terminal client
A coding agent that runs in your shell: it streams from this gateway, reads and edits the project, and asks before it writes or runs anything.
v0.2.10 · x86_64 7.0 MB · static · sha256 234f442f7db72b0f…
curl -fsSL https://panel-ai.neko-build.tech/install.sh | bashBilling
One credit balance funds every request. Prices are per 1M tokens and never rounded at settlement.
Reference
Auth, streaming, the error envelope and both compatibility layers are documented end to end.
Read the API reference