Caedral Python SDK

Official Python client for the Caedral API. OpenAI-compatible request shapes — point your existing code at Caedral with minimal changes.

Installation

pip install caedral

Local development (editable install)

cd sdk-python
python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -e ".[dev]"

Quickstart

from caedral import Caedral

caedral = Caedral(
    api_key="cd_live_...",
    base_url="http://localhost:5001",  # local API gateway
)

completion = caedral.chat.completions.create(
    model="caedral-titan",
    messages=[{"role": "user", "content": "Hello!"}],
)

print(completion.choices[0].message["content"])
caedral.close()

Or use a context manager:

with Caedral(api_key="cd_live_...", base_url="http://localhost:5001") as caedral:
    usage = caedral.usage.get()
    print(usage.weeklyPool.remaining)

Production default base URL: https://api.caedral.com.

Configuration

Parameter	Default	Description
api_key	—	Required. Your cd_live_... API key
base_url	https://api.caedral.com	API gateway base URL
max_retries	3	Retries for idempotent GET requests (exponential backoff)
timeout	120.0	Request timeout in seconds

Methods

caedral.chat.completions.create(...)

OpenAI-compatible chat completions.

Non-streaming:

response = caedral.chat.completions.create(
    model="caedral-olympus",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain quantum computing briefly."},
    ],
    temperature=0.7,
    max_tokens=500,
)

print(response.choices[0].message["content"])
print(response.usage.total_tokens if response.usage else None)

Streaming (generator):

stream = caedral.chat.completions.create(
    model="caedral-titan",
    messages=[{"role": "user", "content": "Write a haiku about code."}],
    stream=True,
)

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

Models: caedral-base, caedral-titan, caedral-olympus, caedral-primordial.

caedral.models.list()

models = caedral.models.list()
for model in models.data:
    print(model.id, model.name, model.pricing_tier)

caedral.usage.get()

usage = caedral.usage.get()
print("Pool remaining:", usage.weeklyPool.remaining)
print("Balance (cents):", usage.balanceCents)
print("Overage used:", usage.overage.usedCents)

caedral.embeddings.create(...)

result = caedral.embeddings.create(
    model="caedral-embed",
    input="Caedral unifies frontier models behind one API.",
)
print(len(result.data[0].embedding))

caedral.images.generate(...)

image = caedral.images.generate(
    model="caedral-vision",
    prompt="A minimal geometric logo on a dark background",
)
print(image.data[0].url or "b64 payload returned")

caedral.audio.generate(...)

audio = caedral.audio.generate(
    model="caedral-voice",
    input="Welcome to Caedral.",
    voice="alloy",
)
print(audio.model)

caedral.rerank.create(...)

ranked = caedral.rerank.create(
    model="caedral-rerank",
    query="billing and subscriptions",
    documents=[
        "Caedral pricing tiers include Starter and Pro.",
        "The API gateway runs on port 5001 in local dev.",
    ],
    top_n=2,
)
for item in ranked.results:
    print(item.index, item.relevance_score)

Error handling

from caedral import Caedral, CaedralAPIError

try:
    caedral.chat.completions.create(
        model="caedral-base",
        messages=[{"role": "user", "content": "Hi"}],
    )
except CaedralAPIError as err:
    print(err.status_code, err.type, err.message)

Async client

AsyncCaedral is planned as a fast-follow. The synchronous client covers all endpoints today.

Integration tests

Requires a running local gateway (http://localhost:5001) and DATABASE_URL in the repo root .env (tests create a temporary API key automatically).

cd sdk-python
pip install -e ".[dev]"
pytest -v

Optional environment variables:

Variable	Description
CAEDRAL_BASE_URL	Gateway URL (default http://localhost:5001)
CAEDRAL_TEST_API_KEY	Skip auto key creation and use an existing key

License

MIT