c3-charm

Python SDK for CHARM, C3 AI's foundation model for multivariate time series.

• Embeddings: turn (T, C) windows into dense vectors for search, clustering, anomaly detection and features.
• Zero-shot forecasting: quantile forecasts and backcasts with no training data.
• Imputation: fill missing values, with quantiles.
• Toolkit (charm_toolkit): anomaly-detection recipes, retrieval metrics, and small trainable heads for forecasting, reconstruction and classification.

The model is channel-aware (channel names are part of the input) and normalizes internally, so you send raw values. It runs server-side; this package is the client. To get an endpoint and key, see Getting access on the model card.

Install

pip install c3-charm                            # core SDK (Python >= 3.10), no torch

# Toolkit (anomaly detection, retrieval, trainable heads) needs PyTorch.
# Install the CPU wheel first, then the toolkit:
pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install "c3-charm[toolkit]"

See installation.md for details.

Quick look

from charm import CharmClient

client = CharmClient(base_url="https://your-charm-server:8080", api_key="...", timeout=300)

fc = client.prediction.create(
    descriptions=[["load_mw", "ambient_temp"]],  # (N, C) channel names
    ts_array=[history],                          # (N, T, C) raw values as nested lists
    target_len=24,                               # > 0 forecast, < 0 backcast
    return_tensors="np",
)
fc.median                     # (N, 24, C) point forecast
fc.denormalized_predictions   # (N, 24, C, Q) quantiles

emb = client.embeddings.create(descriptions=..., ts_array=..., return_tensors="np").embeds  # (N, D)

Using an AI coding agent?

There's a CHARM agent skill that teaches Claude Code, Cursor, Codex and others when and how to apply CHARM, routing them to these docs for the exact API. Install it from https://huggingface.co/c3aiia3c/CHARM.

Documentation

Full documentation ships with the package. Find it with python -c "import charm.docs as d; print(d.get_docs_path())" and start at index.md. It covers concepts and a decision guide, data format, client configuration, embeddings and retrieval, forecasting, anomaly detection, trained heads, and errors plus the raw HTTP API.

License

Apache License 2.0. See LICENSE.