libre-alpha 0.1.1

Liberal Alpha Python SDK for interacting with gRPC-based backend

Liberal Alpha Python SDK

This SDK provides historical upload, historical download, local runner send, and real-time subscription APIs (HTTP + gRPC + WebSocket).

Install

pip install libre-alpha
# Optional: override default API base (default is https://api.librealpha.com)
export LIBALPHA_API_BASE="https://api.librealpha.com"

# Auth (X-API-Key)
export LIBALPHA_API_KEY="YOUR_API_KEY"

# Optional: record encryption key (used to encrypt uploads for encrypted records, and decrypt encrypted payloads)
export LIBALPHA_RECORD_ENCRYPTION_KEY="YOUR_RECORD_ENCRYPTION_KEY"

# Optional: local liberal_runner gRPC address for send_data()
export LIBALPHA_RUNNER_HOST="127.0.0.1"
export LIBALPHA_RUNNER_PORT="8128"

Initialize Client

from liberal_alpha.client import LiberalAlphaClient

# api_base defaults to https://api.librealpha.com
# can be overridden by env LIBALPHA_API_BASE or by passing api_base=...
client = LiberalAlphaClient(
    api_key="YOUR_API_KEY",            # optional if using env LIBALPHA_API_KEY
)

## Historical Upload API (Python) - protobuf stream (/api/entries/history/upload)

This is a **historical backfill** uploader that sends protobuf `DataEntry` messages in a length-prefixed stream:

- Request body format: `[4-byte big-endian length][DataEntry][4-byte length][DataEntry]...`
- Backend groups entries **by minute**; this SDK implementation will also **bucket by minute** and send **1 minute per request**.
- Auth: `X-API-Key` (required).
- Data is carried per-symbol in `symbol_values`:
  - For non-encrypted records: `symbol_values[i].values.items = [float, ...]`
  - For encrypted records: `symbol_values[i].encrypted_payload = <bytes>` (AES-256-GCM)

Client method (recommended):

```python
def upload_data(
    record_id: int,
    df: pd.DataFrame,
    encryption_key: str | None = None,
) -> None:
    ...

• Auth is always taken from client.api_key (or env LIBALPHA_API_KEY when constructing the client).
• encryption_key: record encryption key for encrypting payload when the record is encrypted. If omitted, the SDK uses env LIBALPHA_RECORD_ENCRYPTION_KEY.

Note: the SDK also exposes a module-level upload_data(...) with the same arguments; client.upload_data(...) delegates to it internally. (We only show the signature once here to avoid confusion.)

DataFrame format:

• Required columns:
  • symbol (str): symbol_id string (e.g. "1", "3"). If you pass non-numeric symbol names, the SDK will best-effort map them to symbol_id via /api/records/user-records (target_symbols).
  • timestamp (datetime64 or unix timestamp; will be normalized to UTC and uploaded in microseconds)
• Feature columns (choose ONE approach):
  • Provide a features column containing a list of floats (or a string like "[1.0, nan, 2.0]"), OR
  • Provide user-defined float columns matching the record schema; the SDK will fetch feature order from /api/records/{id}.

Upload packing behavior:

• 1 request = 1 minute (backend requirement)
• Within that minute, the SDK aggregates rows by exact timestamp into 1 DataEntry.
• Each aggregated DataEntry carries multiple symbols in symbol_values: [{symbol_id, values.items}, {symbol_id, values.items}, ...]

Example (upload one minute from a previously-downloaded CSV):

import pandas as pd
from liberal_alpha.client import LiberalAlphaClient

df = pd.read_csv("history_1h_record12_BTC.csv")

# pick the first minute only (backend requires 1 minute per upload)
first_minute = int(df["timestamp"].iloc[0] // 60_000_000)
df1 = df[df["timestamp"].apply(lambda x: int(x // 60_000_000) == first_minute)].copy()
df1 = df1[["symbol", "timestamp", "features"]]

client = LiberalAlphaClient(api_key="YOUR_API_KEY")
client.upload_data(
    record_id=12,
    df=df1,
    # encryption_key="YOUR_RECORD_ENCRYPTION_KEY",  # only needed if record is encrypted (or set LIBALPHA_RECORD_ENCRYPTION_KEY)
)

Historical File Download API (Python) - protobuf (length-prefixed)

def download_historical_data(
    record_id: int,
    symbols: list[str],
    start: datetime | str | int,
    end: datetime | str | int,
    encryption_key: str | None = None,
) -> pandas.DataFrame:
    pass

Parameters:

• record_id: the record id to download
• symbols: list of symbol names (e.g. ["BINANCE_BTCUSDT"]). If you pass [], the SDK will best-effort fetch all symbols from /api/records/user-records (target_symbols).
• start/end: datetime | ISO string | unix timestamp (sec/ms/us)
• encryption_key: optional, record encryption key for decrypting symbol_values[].encrypted_payload

Notes:

• Backend /api/entries/download-links enforces end-start <= 24h (microseconds). The SDK automatically splits long ranges into multiple 24h windows and merges results locally.
• Returned timestamps are in microseconds.
• DataEntry uses repeated SymbolValues symbol_values; each row corresponds to one symbol_id.
• When data is encrypted (symbol_values[].encrypted_payload present), pass encryption_key to decrypt.

Example

from liberal_alpha.client import LiberalAlphaClient
import datetime as dt

client = LiberalAlphaClient(
    api_key="YOUR_API_KEY",
    api_base="https://api.librealpha.com",
)

start = dt.datetime(2026, 1, 1, 0, 0, 0, tzinfo=dt.timezone.utc)
end   = dt.datetime(2026, 1, 3, 0, 0, 0, tzinfo=dt.timezone.utc)  # >24h is OK (SDK will split)

df = client.download_historical_data(
    record_id=2,
    symbols=["BINANCE_BTCUSDT"],
    start=start,
    end=end,
    encryption_key="YOUR_RECORD_ENCRYPTION_KEY",  # for decrypting encrypted data
)

print(df.head())
print("rows:", len(df))

Local Runner Send API (Python) - gRPC ProcessJson

send_data(...) is still available in the Python SDK. It sends one JSON payload to a local liberal_runner through the gRPC JsonService/ProcessJson RPC.

Client method:

def send_data(
    *,
    identifier: str,
    data: dict,
    record_id: int,
    event_type: str = "raw",
    timeout: float | None = None,
):
    ...

Parameters:

• identifier: caller-defined unique id for this payload. The SDK passes it to runner via request metadata.
• data: JSON-serializable payload to send.
• record_id: target backend record id. This is required and is sent in gRPC metadata.
• event_type: runner processor type. Common values are raw and hash. For normal sends, use raw.
• timeout: optional gRPC timeout in seconds. Defaults to the client timeout.

Notes:

• Transport is gRPC, not HTTP.
• The SDK sends to LIBALPHA_RUNNER_HOST:LIBALPHA_RUNNER_PORT (defaults to 127.0.0.1:8128), or to host= / port= passed when constructing LiberalAlphaClient.
• The underlying RPC is JsonService/ProcessJson.
• send_data(...) is the current send entrypoint for local runner usage. There is no separately documented send_alpha(...) API in the current SDK implementation.
• send_data(...) can also send alpha-style payloads. In practice, runner does not distinguish "data" vs "alpha" by RPC method; what matters is that record_id and the fields inside data match the target record schema.
• For alpha records, send the alpha fields directly in data (for example alpha_value) and normally keep event_type="raw".
• If you include symbol in data, runner will use it. If omitted, runner may try to infer it from record configuration.

Example: send normal data to local runner

from liberal_alpha import LiberalAlphaClient

client = LiberalAlphaClient(
    api_key="YOUR_API_KEY",
    host="127.0.0.1",
    port=8128,
)

resp = client.send_data(
    identifier="test_data_001",
    record_id=13,
    event_type="raw",
    data={
        "symbol": "BINANCE_BTCUSDT",
        "Price": 150,
        "Volume": 200,
        "timestamp_ms": 1710000000000,
    },
)

print(resp)

Example: send alpha data to local runner

from liberal_alpha import LiberalAlphaClient

client = LiberalAlphaClient(
    api_key="YOUR_API_KEY",
    host="127.0.0.1",
    port=8128,
)

resp = client.send_data(
    identifier="alpha_001",
    record_id=4,
    event_type="raw",
    data={
        "symbol": "BINANCE_BTCUSDT",
        "alpha_value": 0.1234,
        "timestamp_ms": 1710000000000,
    },
)

print(resp)

Real-time Subscribe API (Python) - WebSocket ws/data

from typing import List, Callable

def subscribe_data(
    record_id: int,
    *,
    symbols: List[str],
    on_data_fn: Callable[[str, pd.DataFrame], None],
    encryption_key: str | None = None,
) -> None:
    """Subscribes to live data for a record into a DataFrame (schema matches publish).

    Args:
        record_id: Backend record identifier.
        symbols: Symbols to filter by (required).
        on_data_fn: Callback function invoked upon receiving new data. The
            function takes two arguments: the symbol (str) and a DataFrame
            containing the new data for that symbol.
        encryption_key: Optional, record encryption key for decrypting encrypted_payload.
    """

Parameters:

• record_id: must be in user's subscriptions (from /api/subscriptions)
• symbols: symbol_id strings to filter by (required). Only rows with symbol_id in this list are passed to on_data_fn.
• on_data_fn: callback invoked upon receiving new data. Takes (symbol: str, df: DataFrame).
• encryption_key: optional, record encryption key for decrypting symbol_values[].encrypted_payload; required if data is encrypted.

Notes:

• Connects to ws/data, sends { wallet_address, record_id } on open.
• If data is encrypted (symbol_values[].encrypted_payload has value), pass encryption_key or set LIBALPHA_RECORD_ENCRYPTION_KEY.
• DataFrame schema matches download_historical_data (record_id, symbol, features/expanded feature columns, timestamps).
• Blocking; run in a daemon thread for long-lived subscription.

Example

import threading
import pandas as pd
from liberal_alpha import LiberalAlphaClient

client = LiberalAlphaClient(api_key="YOUR_API_KEY", api_base="https://api.librealpha.com")

def on_data_fn(symbol: str, df: pd.DataFrame):
    print(f"[{symbol}] entry_id={df['entry_id'].iloc[0]} features_len={len(df['features'].iloc[0])}")

t = threading.Thread(
    target=client.subscribe_data,
    kwargs={"record_id": 2, "symbols": ["1", "3"], "on_data_fn": on_data_fn, "encryption_key": "YOUR_RECORD_ENCRYPTION_KEY"},
    daemon=True,
)
t.start()

print(df.head()) print("rows:", len(df))