DolphinDB

What it is

DolphinDB is a distributed time-series database aimed squarely at quantitative finance, and it is the closest thing in this category to kdb+'s shape: a proprietary engine, a vector language of its own, and as-of joins that were designed in rather than bolted on.

The language is the thing to weigh first. DolphinDB script is a vector and functional language with SQL embedded in it, so aj(trades, quotes, ...) is a function call rather than a keyword — it matches each trade to the last quote at or before its timestamp, and wj generalises that to aggregates over a window around each trade, which is the transaction-cost query written in one line. Both carry a real precondition the docs state plainly: the right table must be sorted on the joining column, and for partitioned tables every joining column but the last must include the partitioning columns.

Streaming is native rather than an ingest protocol. Stream tables publish to subscribers, and engines — including an as-of join engine created with createAsofJoinEngine — compute continuously as rows arrive, so the same join logic runs over live ticks and over history.

Releases are regular: V3.00.6 and V2.00.19 both shipped on 13 July 2026, adding a feature store for machine-learning serving, fixed-income and derivatives pricing functions, JIT compilation improvements and AES-256-CBC encryption. The vendor also sells a family around the core — Swordfish as an embedded library, Shark for GPU acceleration, Starfish as a research platform and DolphinX for LLM agents.

Pricing

A perpetual community licence costs nothing and is capped at 2 nodes, 2 CPU cores and 8 GB of RAM per node. Read that as a development licence: it is a fine place to learn the language, and it is roughly a laptop's worth of resources for a workload whose entire point is not fitting on a laptop.

Above it, nothing is published. There is no pricing page, no per-core rate card and no starting figure anywhere on the vendor's site — an enterprise trial request and a sales conversation is the only route to a number, which is the same position kdb+ occupied before KDB-X.

Data & coverage

None. DolphinDB ships no market data; it stores and computes over ticks you have already sourced.

Integrations

Clients for Python, Java, C++, C#, Go, Rust, Node.js and R, a JDBC driver, an ODBC bridge, a Grafana data source, a VS Code extension and a desktop GUI, most of them Apache-2.0 on GitHub. The dolphindb Python package on PyPI is the vendor's own and reached 3.0.6.0 on 11 August 2026, though its PyPI metadata declares a licence of "DolphinDB" while the client repository it is built from carries Apache-2.0. A vendor-built Excel add-in queries the server from a sheet — navigation pane, editor, and query results or a subscribed stream written straight into cells. The server has been its own MCP endpoint since 3.00.4: addMCPTool, publishMCPTools and callMCPTool turn a DolphinDB function into a tool an LLM client can call, with MCP_EXEC as the access-control right. A separate Stock MCP server in the vendor's GitHub organisation wraps equity-research queries and is documented in Chinese only.

Limitations

• The free tier is not a production tier. Two cores and 8 GB per node is the cap, and the next step up has no published price.
• A second language to learn, and a smaller pool of people who know it than SQL — the closest comparison is q, with the same hiring consequence.
• The server is closed. You cannot read it, patch it or run it past what the licence file permits; the Apache-2.0 repositories are the clients around it.
• English coverage trails the Chinese site. The reference manual, the plugins and the MCP functions are translated; the Stock MCP server is not, and the vendor runs separate Chinese and English tutorial repositories rather than one.
• No market data, no symbology, no corporate actions.

Alternatives

kdb+ is the same bargain with a longer track record and a published price ladder up to 127 GiB. ClickHouse and QuestDB give you ASOF JOIN in SQL under an OSI licence with nothing capped. ArcticDB if the research happens in Python frames and no server is wanted at all.

See the rest of the tick data storage category.