NicoSoft AI documentation

The gateway exposes three universal text endpoints — pick whichever SDK you already use and you get every model. Native pass-throughs (Gemini, image, video) live alongside the universal entries when you need provider-exact shapes.

• Drop-in for the OpenAI, Anthropic, and Responses SDKs.
• Pay-as-you-go on the same balance, regardless of upstream provider.
• Automatic channel failover when an upstream returns 401 / 403 / 5xx.
• Free-tier models routed through pooled keys so you can try them with zero balance.

Make your first request in under a minute. The OpenAI Python SDK works as-is — only the base_url and api_key change.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.nicosoft.ai/v1",
    api_key="sk-nsai-...",
)

resp = client.chat.completions.create(
    model="anthropic/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Hi"}],
    stream=True,
)

for chunk in resp:
    print(chunk.choices[0].delta.content or "", end="")

All requests require a bearer token. Keys are scoped to a single account and can carry independent credit caps and per-minute limits.

Three universal entries accept any text model. Native pass-throughs preserve provider-exact shapes for clients that need them.

Drop-in replacement for the official OpenAI API. Accepts every text model — OpenAI, Anthropic, Gemini, DeepSeek, Azure — by switching the model slug.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.nicosoft.ai/v1",
    api_key="sk-nsai-...",
)

r = client.chat.completions.create(
    model="openai/gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

Drop-in for the Anthropic SDK. Set the SDK's base_url to https://api.nicosoft.ai and call messages.create. Non-Anthropic models are bridged — request, response, and streaming are all translated.

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.nicosoft.ai",
    api_key="sk-nsai-...",
)

msg = client.messages.create(
    model="anthropic/claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hi"}],
)
print(msg.content[0].text)

Gemini-native endpoints are a strict 1:1 pass-through — point Google's Gen AI SDK at the base URL and the request, response, and stream all preserve Google's original schema. If you'd rather treat Gemini as a regular OpenAI / Anthropic model, send it through /v1/chat/completions or /v1/messages instead.

Every endpoint supports SSE streaming. Set "stream": true (OpenAI / Anthropic families) or use :streamGenerateContent (Gemini).

Pass toolson the request and the gateway forwards the schema. Tool-call output is normalised into the request's protocol shape — your existing OpenAI / Anthropic tool-handling loop works regardless of which model actually runs.

Concurrent / parallel tool calls and tool_choice constraints (auto, required, named) pass through unchanged.

OpenAI-compatible synchronous image generation. The OpenAI SDK works directly — set base_url to https://api.nicosoft.ai/v1 and call client.images.generate(...). Billing is per-image; failed requests aren't charged.

Google Imagen is exposed via the Gemini native endpoint at /v1beta/models/imagen-*:generate.

Model slugs are namespaced by upstream family so you can tell at a glance where a request is going.

• openai/* — OpenAI native (GPT-4o, GPT-5, etc.)
• anthropic/* — Anthropic native (Claude Sonnet / Opus / Haiku)
• gemini-* / google/* — Google Gemini family
• deepseek/* — DeepSeek
• azure/* — Azure-hosted OpenAI models
• nicosoft/* — first-party hosted variants

Browse the full catalog at /models.

On error the gateway returns a JSON envelope with success: falseand a machine-readable code. Model provider errors are normalised so provider internals aren't exposed to clients.

Rate limits are enforced per API key using a sliding window. When you exceed the limit the gateway returns 429 RATE_LIMITED with a Retry-After header (seconds until the window resets).

Defaults are generous for typical usage. For high- throughput or batch workloads, get in touch about dedicated quotas.

Questions or issues — support@nicosoft.dev. Status page: /status. Model catalog: /models.