Ogun, a local-first workflow platform for data engineers who need repeatable datasets
Alpha 0.1.0-alpha.1 · published 29 October 2026. Windows and Linux desktop app. macOS, Android and iOS are planned.
Ogun is built to turn web and file sources into datasets you can rebuild and trace. Agents may plan the work, and a person approves every step that runs code.
Last reviewed
What works today and what is not there yet
Works today
- Ogun Studio, a desktop app for Windows and Linux that starts its own engine and needs no account or network.
- Workflows stored as hashed JSON documents, with runs, schedules, partitions and backfills that resume only unfinished work.
- Code operators that run in a sandbox only after a person approves their exact hash.
- Exports to a folder, a SQLite file or an S3-compatible bucket.
- An optional agent that uses your own provider key or an installed agent CLI, bounded by a closed tool table, spend caps and an offline switch.
Not there yet
- Connectors checked against the real services. Most connectors are descriptions only.
- A call to a live model provider with a real key. The agent was checked against local test servers.
- Remote approvals, which stay off until an external security review.
- macOS, Android and iOS builds, and signed Windows binaries.
What Ogun is built to do
- A long collection job dies and you start again.
- Built to resume partitions and backfills after a crash, redoing only the unfinished work.
- Nobody can say how a dataset was made.
- Designed to store results by content and record lineage for every run.
- An agent writes a shell command and you run it.
- Designed so an agent proposes a workflow as data. A validator checks it, a plan diff shows the change, and a person approves.
For your engineers
- The ogund daemon stores workflows as JSON with a canonical hash and emits OpenLineage events.
- Artifacts sit in a content-addressed store. Approvals are bound to a request hash.
- Code operators run in a Windows AppContainer or bubblewrap on Linux, with pinned versions.
- The CLI covers validate, hash, diff, run, backfill, approvals, ogun mcp and ogun lsp.
- It does not replace Airflow, a dataframe library, a SQL engine, an OCR engine, a vector database or an RPA tool.
- The builds are unsigned, so Windows SmartScreen warns before the first run.
Ask for an Ogun walkthrough
We run a workflow on a desktop build, show its approvals and lineage, and name what the alpha does not do yet.