inertialai-python 0.1.0

The official Python client for the InertialAI API — generate multi-modal embeddings from time-series and text.

inertialai-python

The official Python client for the InertialAI API.

InertialAI builds foundation models for time-series and real-world sensor data. This library wraps the InertialAI REST API and gives Python developers a typed, ergonomic way to generate multi-modal embeddings from time-series signals, text, or both.

Beta. The API surface is stable enough for real use but may change before 1.0. Pin to a specific version in production.

Installation

pip install inertialai-python

Requires Python 3.11 or newer.

Authentication

Create an API key in the InertialAI app and expose it as an environment variable:

export INERTIALAI_API_KEY="iai_..."

The client reads INERTIALAI_API_KEY automatically. You can also pass it explicitly via the api_key constructor argument.

Quickstart

Synchronous

from inertialai_python import EmbeddingModel, InertialAI
from inertialai_python.types import EmbeddingMultiModalInput

client = InertialAI()

response = client.embeddings.create(
    model=EmbeddingModel.INERTIAL_EMBED_ALPHA,
    input=[
        EmbeddingMultiModalInput(
            time_series=[
                [0.01, 0.02, 0.04, 0.03, 0.02],  # channel 1 (e.g. accel x)
                [0.00, 0.01, 0.01, 0.02, 0.01],  # channel 2 (e.g. accel y)
            ],
            text="walking, 50 Hz accelerometer",
        ),
    ],
)

vector = response.data[0].embedding
print(f"Got a {len(vector)}-dim embedding using {response.model}")

Asynchronous

import asyncio

from inertialai_python import AsyncInertialAI, EmbeddingModel
from inertialai_python.types import EmbeddingMultiModalInput

async def main() -> None:
    async with AsyncInertialAI() as client:
        response = await client.embeddings.create(
            model=EmbeddingModel.INERTIAL_EMBED_ALPHA,
            input=[EmbeddingMultiModalInput(text="sensor reading from accelerometer")],
        )
        print(response.data[0].embedding[:8])

asyncio.run(main())

Multi-modal input

Each EmbeddingMultiModalInput must contain at least one of:

Field	Type	Description
time_series	list[list[float]]	Multi-channel signal. Outer list = channels, inner list = samples per channel.
text	str	Natural-language description to embed alongside (or instead of) the signal.

You can batch many inputs in a single request by passing a list. Each returned EmbeddingData carries the index of its corresponding input.

Configuring the client

from inertialai_python import InertialAI

client = InertialAI(
    api_key="iai_...",           # overrides INERTIALAI_API_KEY
    base_url="https://inertialai.com",
    timeout=30.0,               # seconds, or an httpx.Timeout instance
    max_retries=2,              # retries for transient errors (429, 5xx, network)
)

Both InertialAI and AsyncInertialAI support context-manager usage to ensure the underlying HTTP connection is closed:

with InertialAI() as client:
    ...

async with AsyncInertialAI() as client:
    ...

Error handling

All API errors derive from InertialAIError:

InertialAIError
├── APIError
│   └── APIStatusError
│       ├── AuthenticationError   (401)
│       ├── ValidationError       (422)
│       ├── RateLimitError        (429)
│       └── InternalServerError   (5xx)
└── APIConnectionError
    └── APITimeoutError

from inertialai_python import (
    AuthenticationError,
    InertialAI,
    RateLimitError,
    ValidationError,
)

client = InertialAI()

try:
    client.embeddings.create(model="inertial-embed-alpha", input=[...])
except AuthenticationError:
    print("Check your INERTIALAI_API_KEY")
except RateLimitError:
    print("Slow down — you've been rate limited")
except ValidationError as e:
    print(f"Invalid request: {e}")

The client automatically retries transient failures (429, 500, 502, 503, 504, network errors, timeouts) with exponential backoff and jitter, honoring the Retry-After header when present.

OpenAI SDK compatibility

The InertialAI embeddings endpoint is interface-compatible with OpenAI's /v1/embeddings API, so the openai Python client will work against it if you json.dumps multi-modal inputs by hand. This library exists to give you a first-class, typed experience — no manual serialization, native EmbeddingMultiModalInput models, and typed error hierarchy.

Links

• Website: https://www.inertialai.com
• Documentation: https://docs.inertialai.com
• Embeddings guide: https://docs.inertialai.com/docs/using-the-embeddings-endpoint
• Dashboard / API keys: https://app.inertialai.com
• Issues: https://github.com/InertialAI/inertialai-python/issues
• Changelog: CHANGELOG.md
• Contributing: CONTRIBUTING.md

License

Apache 2.0 — see LICENSE.