DuckDB 2.0 Ships Server Mode, Agent CLI

// 1h agoOPENSOURCE RELEASE

DuckDB 2.0 Ships Server Mode, Agent CLI

DuckDB 2.0.0 “Cinnamon Teal” is a major LTS release adding Quack client-server mode, asynchronous I/O, first-class VARIANT data, triggers, an extensible SQL parser, and agent-friendly CLI output. It is supported through at least October 2027.

// ANALYSIS

DuckDB is evolving from a lightweight embedded analytics engine into a broader data platform for local, remote, and AI-assisted workflows.

–Agent mode reduces CLI output by 59% in the team’s TPC-H tests and reports errors as JSON, making DuckDB easier for coding agents to operate.
–Asynchronous object-storage I/O and `NEAREST` joins strengthen DuckDB for Parquet, vector, and embedding workloads.
–Quack gives DuckDB a server option without abandoning its embedded-first design.
–The new storage format is not backward-readable by pre-2.0 versions, so production upgrades require shadow testing and migration planning.

// TAGS

duckdbopen-sourcedata-toolsagentvector-dbself-hosted

DISCOVERED

1h ago

2026-10-21

PUBLISHED

4h ago

2026-10-21

RELEVANCE

7/ 10

AUTHOR

nnx

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH53m ago

ChatGPT Slides puts teams inside deck creation

ChatGPT Slides turns conversations, files, outlines, or templates into editable presentations that teammates can revise alongside ChatGPT. Users can present within ChatGPT or export decks to PowerPoint, PDF, or Google Slides.

MODEL1h ago

Flint drops 40B open coding model

Recursive released Flint, a 39.8B-parameter mixture-of-experts coding model with roughly 2.5B active parameters per token, a 262K context window, and Apache 2.0 open weights. It is designed for local coding agents with reasoning and multi-turn tool use.

MODEL1h ago

Recursive Flint drops AI-built coding model

Recursive’s first model, Flint, is a 39.8B sparse MoE coding model with roughly 2.5B active parameters, a 262K-token context window, tool use, and Apache 2.0 weights. The announcement claims 125+ tokens per second on short-context MacBook workloads, enabling private local coding without per-token API fees.