Native-speed ETL. Run it on your server or in your cloud.
Odara ETL is a data-integration platform written in Rust. Design pipelines visually with SQL and Python, then run them Local, in Docker, on AWS ECS Fargate or on Google Cloud Run.
v0.4.4 · Linux (.deb, .rpm, tar.gz) and Windows · Proprietary, free to download
Works with
- Rust
- Python
- Apache Arrow
- Apache DataFusion
- PostgreSQL
- MySQL
- Oracle
- SQL Server
- MongoDB
- Snowflake
- BigQuery
- Redshift
- ClickHouse
- DuckDB
- SQLite
- Amazon S3
- Azure Blob Storage
- Google Drive
- Docker
- AWS Fargate
- Google Cloud Run
One pipeline. Each environment decides where it runs.
DEV, STG and PROD hold values. Where each one runs is a setting of the environment: Local, Docker, AWS ECS Fargate or Google Cloud Run. AWS and Google Cloud support ship in every download.
Execution setting of STGDocker
Odara server
Coordinates only. It holds least-privilege rights: start, watch and stop its runs, read their logs. It never holds the data.
The editor, the API, schedules and Maestro all start runs here.
Run package
The pipeline and its code. No secret values inside.
Per-run vault
Created for the run with its secrets, deleted when the run ends.
LocalDEV
Inside the Odara server, as always. The default while you build.
Choose the environment of a runDockerSTG
One container per run on the server, with CPU, memory and time limits.
Set up Docker, step by stepAWS ECS FargatePROD
One task per run in your AWS account. Deleted when the run ends.
Set up AWS ECS Fargate, step by stepGoogle Cloud Run
One job execution per run in your Google Cloud project. Deleted when the run ends.
Set up Google Cloud Run, step by step
- Secrets stay home
- Secret values never travel in the run package. On AWS and Google, each run gets its own vault entry, deleted when the run ends.
- Least privilege
- The Odara server only gets the rights to start, watch and stop its runs and to read their logs.
- Same Monitor
- An isolated run is recorded like a local one: per-node rows and times, the target history, and the environment it ran in.
- Limits that hold
- Abort and the time limit stop the run. Out of memory is reported as such and never touches the server.
Validated end to end on real AWS and real Google Cloud. Setup takes minutes: a ready CloudFormation template for AWS and a setup script for Google Cloud. Step-by-step guides with screenshots: Docker · AWS ECS Fargate · Google Cloud Run.
Three things you can check on the first day.
- 01
One native binary. No JVM.
The engine, the API and the editor ship as a single executable. There is no runtime to install, size or tune.
- 02
Installed in about a minute.
One command on Linux (.deb, .rpm, tar.gz) or Windows (PowerShell). Open the browser and start building.
- 03
SQL and Python in the same pipeline.
SQL runs on Apache DataFusion. A Python step sits next to it. Data moves between nodes as Apache Arrow batches, never converted to rows or text.
Bring what you already have.
A Talend project import you review before anything is built, and 70 node types for the rest of your stack.
Talend Import Utility Beta
- 1Upload the project export
The .zip from Talend Studio, or the project folder zipped from git. X-Ray lists every job and component, groups similar jobs, and sizes the effort.
- 2Understand, then approve
The AI, with your key, explains what each job does: purpose, steps, business rules, checks, open questions. You correct it and approve it.
- 3Import
Jobs become Odara pipelines; jobs that only call other jobs become Maestros. Anything unsupported is reported, not guessed.
tDBInputPostgreSQL ReadtMapMappertFileInputDelimitedCSV SourcetJavaRowPythontAggregateRowSQL (DataFusion)tRunJobMaestro
70node types in 0.4.0
Counted from the source: 66 palette nodes, the Mapper, 3 Maestro nodes.
- PostgreSQL
- Oracle
- SQL Server
- MySQL
- Snowflake
- BigQuery
- Amazon S3
- Azure Blob Storage
- Google Drive
- Excel
- Python
- Docker
Three steps to a pipeline in production.
Install, design on the canvas or ask the assistant, then choose where each environment runs. The pipeline never changes.
- 1
Install
One command on Linux (.deb, .rpm, tar.gz) or Windows (PowerShell). Leave your name and email below and the command appears.
Get the install command - 2
Design, or ask the AI
Place nodes on the canvas and map columns in the Mapper. Or describe the pipeline in a sentence: the assistant writes the SQL, the Python and the nodes. Bring your own key: Anthropic Claude, Gemini, DeepSeek or any OpenAI-compatible API.
- 3
Choose where it runs
In Admin › Environments, set each environment’s Execution: Local, Docker, AWS ECS Fargate or Google Cloud Run. In the editor, pick the environment of a run from the top bar; schedules and Maestro follow the same setting.
Early users, verbatim.
Most ETL platforms force you to choose between ease of use and computational power. Odara breaks that paradigm. By pairing a blazing-fast, Arrow-native Rust engine with an AI assistant that actually understands data schemas without hallucinating, it eliminates friction and accelerates delivery. Robustly engineered, it is exactly what the modern data stack has been waiting for.

Odara helped me a lot on a data-integration project. Installation and setup were simple, and I was able to build and ship pipelines to production quickly — even without any prior training. The tool is intuitive and practical, and the AI support makes creating and maintaining the processes even easier.

One node, any format — CSV, JSON, Parquet, Excel, even a file with no extension. Magic File node quietly deleted the most tedious part of every pipeline I build.

Odara is a real game changer in the ETL world, empowered by AI, easy to use. It delivers immediate value from day one.

Download Odara ETL 0.4.4.
Linux (.deb, .rpm, tar.gz) and Windows. AWS and Google Cloud support in every download. Free.
- Ubuntu 22.04+, Debian 12+, Fedora 36+ · x86_64
- Windows 10 and 11 (64-bit) · PowerShell 5.1+
- Installs to
~/.local/share/odaraor%LOCALAPPDATA%\Programs\Odara