Changelog
Noteworthy milestones from across the Hotdata platform.
July 2026
Fork a database
Instant server-side copy of a managed database — branch your data like you branch code.
Mutable managed tables
Append, upsert, and keyed deletes, plus Chain to materialize query results into tables.
Concurrency benchmark published
Query latency stays flat from 1 to 50 readers, compared against Snowflake and BigQuery. Read more →
June 2026
datafusion-ducklake open-sourced
DuckLake support for Apache DataFusion, donated to datafusion-contrib; now Hotdata's storage backend. Read more →
Rust SDK published
The
hotdatacrate, with Arrow decoding and resumable uploads.Direct-to-storage uploads
Files stream in resumable chunks, so any size table loads from any machine.
May 2026
Managed databases launch
Create in under a second, load Parquet, query; auto-expiry makes throwaway databases practical.
The integration wave
Ibis, marimo, LangChain, LlamaIndex, Jupyter, and the dlt destination.
Arrow-native results
Results stream as Arrow IPC, plus CSV, Markdown, and Parquet formats.
April 2026
Semantic search with automatic embeddings
vector_search()embeds your query text for you. Read more →Python SDK published
pip install hotdata.
March 2026
The Hotdata CLI launches
Auth, workspaces, and SQL from the terminal, via Homebrew.
Full-text and vector search
BM25 and HNSW indexes, queryable in plain SQL.
Public API, formalized
OpenAPI spec with browsable reference docs.
February 2026
hotdata.dev goes live
Company site, docs, and blog.
Geospatial SQL
PostGIS-style
ST_*functions and GeoParquet output.
January 2026
Smart caching layer lands in the engine
The foundation of Hotdata's sub-second queries.
Query results become data
Every result is saved and queryable with SQL.