The open time‑series
lakehouse

Hover the chart to write data, and the tabs on the right to pick a reader.

Features

Fast, cheap time-series queries on open files in your own bucket, with no server to run.

Fast

Queries on S3 return in 133 ms, and are 6⁠–⁠18× faster than other lakes on fresh data.

Benchmarks

Cheap

You pay only for storage, reads, and writes. Data costs about 6× less to keep than in TimescaleDB.

Cost

Analytics

An optimized toolkit for analyzing time series, built into the engine and storage.

Performance

Open

Standard Iceberg and Parquet in your own bucket, so you’re not locked in.

Querying

Serverless

No database or maintenance jobs to run. Timelake is a library that runs in your process.

How it works

Any object store

Runs on AWS S3, Google Cloud Storage, Azure, R2, MinIO, or your local disk.

Storage

Data

Time series from machines, vehicles, markets, and servers.

Industrial IoT

Thousands of sensor tags sampled every second to every minute, with backlogs sent after outages.

Vehicle fleets

Position and diagnostics from thousands of vehicles and devices, every few seconds.

Robotics

Joint positions, torques, and sensor readings from robots, recorded at high rates and replayed by time range.

Financial data

Prices and trades for thousands of instruments, read as time ranges, bars, and the latest price.

Server metrics

CPU, memory, and disk metrics from thousands of hosts.

Uses

Read straight from your bucket by apps, notebooks, functions, and agents.

Dashboards and apps

Charts and apps that read the bucket over HTTPS with the TypeScript client, in the browser or at the edge.

Data science and ML

Time series pulled into notebooks as Arrow tables, for pandas, polars, and model training, straight from the bucket.

Serverless functions

Lambda functions and short jobs that start, append or query, and exit. There are no connections to pool and no database to size.

AI agents

Agents can query as often as they like: reads go to object storage, so there is no database for them to overload.

Architecture

From the sources that send data to the tools that read it, with only your bucket in between.

Field device
Sensor
Market data
Feed
Robot arm
Robot
Backfill
Files
IngestRules · planned
Kinesis
Events · planned
Timelakein your process
Use from
Python · Rust · CLI · TS
Read
SQL on DataFusion
Write
Sort · compact · index
Your bucketfiles only
Storage
S3 · GCS · Azure · R2
Tables
Apache Iceberg
Files
Parquet
Catalog
Files, no server
AWS Lambda
Function
AI agent
AI · planned
Dashboard
App
DuckDB · Spark PyIceberg
Engines