A real recording of the cne serve dashboard.

Why CNEquity

How it differs from common data tools (full write-up in the comparison):

	AKShare	Tushare Pro	Baostock	CNEquity
Account and token	Not needed	Registration + token, unlocked by points	Not needed, anonymous login	Not needed for basic collection
Storage and increments	Caller's job	Caller's job	Caller's job	Local Parquet with built-in increments and resume
Delisted stocks and PIT financials	Depends on the interface	Depends on the interface	Partly provided	Delisted stocks kept; financials strictly PIT by announcement date
Row-level provenance	None	None	None	Every row carries source, version and fetch time
Use by AI	Wrap it yourself	Wrap it yourself	Wrap it yourself	Read-only via cne mcp

Quick start

Requires Python 3.10+ on macOS, Linux or Windows. No account or token needed.

① Try it in a minute. Fetch the last 30 trading days for 5 stocks from live sources into a separate directory that never touches your real lake:

pip install cnequity
cne init --profile demo   # about a minute; prints the bars, a sample query and the command to hand it to an AI agent
cne serve --config configs/cnequity.demo.toml   # open the dashboard shown above

② Let an AI agent query A-shares. Connect the lake read-only to Claude, ChatGPT, Codex, Gemini CLI, Cursor, VS Code or any other MCP client (give --config as an absolute path). With Claude Code:

claude mcp add cnequity -- cne mcp --config /abs/path/to/configs/cnequity.demo.toml

See the MCP guide for every client; ChatGPT only reaches public URLs, so it connects through cne mcp --http --token and a tunnel. Then ask: "Which day had CATL's highest close in the last 30 trading days?"

Real MCP tool calls and results on the demo lake; closes are unadjusted.

③ Build the full lake. Every Shanghai, Shenzhen and Beijing A-share, last 3 years:

cne init     # the first run can take hours; if it stops, run it again and nothing already downloaded is fetched twice
             # a few-second network preflight runs first; a failure in a non-critical step such as corporate actions does not stop the build, and the summary lists what is left
cne check    # confirm the data is complete and usable

from cnequity.query import load

bars = load("daily_bars", symbols=["600519.SH"])
print(bars.tail())

Then run cne run daily once a day to stay current. For longer history, or what init does step by step, see the initialization guide.

What can you research?

Question	Data and example
Returns across dividends and splits	daily_bars + adj_factors · adjustment recipe
Financial facts visible on a rebalance date	financial_statement_items · PIT recipe
Historical universes and pre-delisting prices	instruments, trading_status, delisting_events · profiles
Valuation, flows and sector rotation	Valuation, capital and structure datasets · query guide
Futures curves, option chains and Greeks	Contract-level prices and derived datasets · derivatives

Data coverage

57 datasets (49 curated, 8 derived), organized into L0–L9. Fields, keys, history starts and sources are in the catalog and source notes.

Tier	Research use	Representative datasets
L0	Reference	Instruments, trading calendar, trading status
L1	Market data	Daily and index bars, adjustment factors, optional minutes and trade snapshots, delisting events
L2	Corporate events	Actions, announcement index, disclosure schedule
L3	Fundamentals	Financials, valuations, share structure, holders, consensus
L4	Capital and flows	Northbound, margin trading, dragon-tiger lists, block trades, fund flows
L5	Industry structure	Index constituents, industry and sector membership, industry index
L6	Macro	Indicators and market breadth
L7	News and rotation	News, wires, sentiment, rankings, sector prices and flows
L8	Risk and regulation	Share unlocks and regulatory events
L9	Derivatives	Futures and options contracts, contract bars, continuous futures, Greeks and minutes

Minutes, trade snapshots, futures/options, and best-effort datasets such as money flow, news and popularity ranks are off by default. Turn them on when you need them.

Architecture

Survivorship bias

Today's roster is not a historical universe.

Daily updates and upgrades

cne schedule install   # install a daily job for the current user (macOS / Linux / Windows)
cne run daily          # or run this yourself once a day
cne check              # acceptance; exits non-zero on gaps or quality errors

To upgrade: pip install -U cnequity, then cne config upgrade and cne migrate (preview first, then add --apply). Scheduling, the operations page and failure follow-up are in the runbook; scope and resume in initialization.

Fit and limits

CNEquity fits ongoing history, repeated queries and self-hosted data.

The project is in 0.x, with no backtesting engine or order execution. Code is Apache-2.0; upstream data has separate terms. No data lake is distributed with the repository.

Documentation and contributions

Task	Read
Install and get a result	Installation · Quickstart
FAQ	Free A-share historical data · Avoiding look-ahead bias · Ask AI about A-shares · AKShare, Tushare or Baostock
Understand the data	Catalog · Schemas · Recipes
Find a command	CLI · Options · Network/write effects
Operate the lake	Runbook · Fetch policy · Troubleshooting
Product direction and feedback	Product overview · Upgrades and feedback

Issues with minimal reproductions, documentation fixes and adapter PRs are welcome. Cite CITATION.cff for research; report vulnerabilities privately using the security policy.

Useful to you? Star CNEquity, or share it with someone maintaining China-market data.