brokerkit-core

The framework half of BrokerKit: the models, the interfaces every broker adapter implements, and the assembly layer that resolves create_broker("zerodha", ...) to a live, authenticated broker.

This package ships no broker integrations. Install it alongside at least one adapter.

pip install brokerkit-core brokerkit-zerodha

Available adapters: brokerkit-groww, brokerkit-fyers, brokerkit-upstox, brokerkit-dhan, brokerkit-angelone, brokerkit-zerodha.

The idea

Every Indian broker ships its own SDK with its own auth flow, symbol format, units (some report prices in paise, some in rupees), error convention (some raise, some return a failure dict, some return status: false and only log it) and its own idea of what a "quote" contains. brokerkit-core defines one typed, async contract; each adapter absorbs its broker's quirks behind it.

from brokerkit import create_broker

broker = await create_broker("zerodha", api_key=..., api_secret=..., access_token=...)

instruments = await broker.instruments.fetch_instruments()
reliance = next(i for i in instruments if i.symbol == "RELIANCE")

quote = await broker.market.get_quote(reliance)
print(quote.last_price, quote.volume)

await broker.close()

Swap "zerodha" for "fyers" and nothing else changes.

The contract

Every broker exposes the same providers:

Plus three interfaces that only some brokers implement, deliberately kept off the shared Broker base so no broker is forced to carry a hole: ChargesProvider, FundamentalsProvider, NewsProvider, MarketInformationProvider.

Design decisions worth knowing

• Async-native. Every interface method is async def. Adapters wrap sync vendor SDKs with asyncio.to_thread; blocking calls never run inside an interface method.
• Pydantic v2 models, Decimal prices. Validation at construction, JSON-serializable for API/agent layers. Enums are stdlib StrEnum.
• Instruments are a thin adapter, not a store. fetch_instruments() downloads, normalizes and returns — the framework holds no cache, no indexes, no lookups. Storage and querying belong to your app. Calling it again re-downloads, because brokers update masters daily.
• Instance-scoped state only. No module-level tokens, clients or caches, so one process can hold several accounts. BrokerManager composes N named brokers on top of that.
• Adapters are plugins. Core never imports an adapter. create_broker("dhan", ...) lazily imports brokerkit_dhan, whose Broker subclass self-registers via __init_subclass__. Adding a broker touches no core code.

Models

Instrument, Order / OrderRequest, Position, Holding, Quote / Ohlc / DepthLevel, Candle, Tick, OptionChain / OptionContract / OptionGreeks, BrokerageCharges, and the fundamentals/news/market-information models.

Two cross-broker facts these encode:

• Tick.minute_ohlc only ever populates for Upstox — it is the one broker whose feed computes a live 1-minute candle server-side. Everywhere else it stays None and you aggregate ticks yourself.
• Instrument.isin is None for Angel One and Zerodha — their instrument masters have no ISIN column at all. ISIN is the natural cross-broker join key for equities, so join those two on exchange_token instead; it matches exactly across Groww, Fyers, Dhan and Zerodha.

Multiple accounts

from brokerkit import BrokerManager

manager = BrokerManager()
await manager.add("main", "zerodha", api_key=..., api_secret=..., access_token=...)
await manager.add("data", "fyers", client_id=..., secret_key=..., ...)

quote = await manager["data"].market.get_quote(instrument)
await manager.close_all()

License

MIT © 2026 Aditya Vishwakarma