PROJECT DIRECTION / UPDATED 11 OCTOBER 2026

Make data collection
easier to build and maintain.

The aim is a connected workflow: understand a source, write its extractor, test the records, release the change and see what happens when it runs again.

The next priorities are easier Enterprise installation and assistance with writing and repairing extraction code. This roadmap describes the direction; delivery dates will follow as releases take shape.

Working foundation

The edit, test and correct loop

Local automation

Complete and standalone Node have public Windows alpha downloads. Complete brings together a visible browser, scripts, workspaces, queues and MCP tools.

Read Windows release notes →

Enterprise Studio

Edit sources, run HTTP/browser draft tests, correct errors and reuse samples. Shape record fields and mappings in the same workspace, then prepare the product for review and release.

Follow the Studio workflow →

Distribution building blocks

Seven Enterprise Linux container images are publicly available. Matching Windows packages provide draft execution and managed browser hosts for guided evaluations.

Understand the published images →
Immediate priority

Make the first evaluation reproducible

The next distribution step is a matching installation bundle with pull-only Compose configuration, setup utilities, generated installation credentials, an example workspace and a clear operator handoff.

The aim is to let a new tester get from downloaded packages to a first successful draft with clear instructions. Until that bundle is available, evaluations use a prepared installation with operator support.

See the current setup path →
Experimental / next integration step

Help write and repair the extractor

Two entry points share the draft workflow: an external assistant through Enterprise authoring MCP, and an assistant inside Studio. Both are designed around saved source context, reviewable changes and draft tests.

  1. Provide context

    Use the saved source, desired record contract and relevant source evidence.

  2. Review changes

    Inspect proposed code and decide what to save.

  3. Run the draft

    Test the saved implementation with an explicit input.

  4. Correct a failure

    Use the actual error and result to guide the next edit.

AI assistance is optional and experimental. The next step is to validate the complete generation and repair workflow with live providers. See the assistant guide for setup and current integration status.

Provider order

First target

LiteLLM

Establish the complete Studio generation and repair loop through a configured gateway.

Retained direction

Ollama / local inference

Keep a local-model option, with practical guidance on model choice and hardware requirements for extraction tasks.

Near-term follow-up

OpenRouter

Follow the initial gateway/local workflow with provider-specific verification and documented setup.

Longer-term direction

Broader automation and operational confidence

Extend authoring help across the data-product workflow and make larger installations easier to operate. Future work includes documented upgrade and recovery paths, fleet management and testing against demanding workloads. The current alpha focuses on guided evaluation.

Alpha availabilityAbout the independent projectChoose your first task →