API reference

Quickstart

Three steps. Get an API key from your dashboard, set it as an environment variable, send a request.

1. 1 Sign up at cloud-horizons.com/ai and generate a key.
2. 2 Export it: export CLOUD_HORIZONS_KEY=ch-...
3. 3 Send a request:

curl https://api.cloudhorizons.ai/v1/chat/completions \
  -H "Authorization: Bearer $CLOUD_HORIZONS_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k2.5",
    "messages": [
      {"role": "user", "content": "Hello, in two lines."}
    ]
  }'

Authentication

Bearer token in the Authorization header. Same shape as OpenAI.

Authorization: Bearer ch-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Keys are scoped per project. Rotate from the dashboard, old keys are revoked instantly. The team plan supports SSO and per-seat keys.

Endpoints

Stable v1 surface. We mirror the OpenAI Chat Completions, Embeddings, and Models endpoints. The Responses API arrives later in 2026.

Models

Pass the model slug in the request body. Aliases like latest-reasoning and latest-coder point at the current best model in each category.

• kimi-k2.5 — long-doc reasoning, 200K context
• minimax-m2.5 — multilingual chat, agent loops
• glm-4.6 — tool use, structured output
• qwen-3-coder — pure code generation
• llama-3.3-70b — general-purpose chat
• mistral-large-3 — European language fluency
• bge-m3 — embeddings, 1024 dims

Streaming

Set stream: true and we emit Server-Sent Events with the same delta shape OpenAI uses. The Python and Node SDKs handle the parsing for you.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cloudhorizons.ai/v1",
    api_key="ch-...",
)

stream = client.chat.completions.create(
    model="glm-4.6",
    messages=[{"role": "user", "content": "Stream a haiku"}],
    stream=True,
)

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

Tool calling

Define tools as JSON schemas, the model returns a tool call when it needs one. Same surface as OpenAI's tools parameter.

tools = [{
    "type": "function",
    "function": {
        "name": "get_alerts",
        "description": "Fetch active PagerDuty incidents",
        "parameters": {
            "type": "object",
            "properties": {
                "service_id": {"type": "string"},
                "since": {"type": "string", "format": "date-time"},
            },
            "required": ["service_id"],
        },
    },
}]

resp = client.chat.completions.create(
    model="glm-4.6",
    messages=[{"role": "user", "content": "Any active alerts on api-gateway since 1pm?"}],
    tools=tools,
)

call = resp.choices[0].message.tool_calls[0]
print(call.function.name, call.function.arguments)

Embeddings

BGE-M3 by default. 1024 dimensions, multilingual, strong on retrieval. Pass a list of strings, get a list of vectors back.

resp = client.embeddings.create(
    model="bge-m3",
    input=["EU AI Act compliance", "Schrems II ruling"],
)

for emb in resp.data:
    print(len(emb.embedding))  # 1024 dimensions

Errors

JSON body, OpenAI-shaped. Type field is the stable contract, message is human-readable and may change.

{
  "error": {
    "type": "rate_limited",
    "message": "Per-minute token limit reached. Retry after 12s.",
    "retry_after_seconds": 12
  }
}

Rate limits

Limits are per project. We return 429 with a retry_after_seconds field, the SDKs honor it automatically.

Data handling

Your prompts and completions are not used to train any model. Inference happens in EU regions only. Logs are kept for debugging and abuse detection, default 7 days, configurable to 0 days on the team plan. Bring your own key for total isolation.

Standard EU Data Processing Agreement available on team plans. Sub-processors disclosed at /ai/changelog.

SDKs

We do not ship our own SDK. The OpenAI clients work as-is, override the base URL.

• Python: openai v1+, set base_url
• Node: openai v4+, set baseURL
• Go: openai-go with custom option.WithBaseURL
• LangChain: ChatOpenAI(base_url=...)
• LiteLLM: route as openai/<model> with api_base
• Vercel AI SDK: custom OpenAI provider, set baseURL