Skip to content

Project Roadmap

📅 Overview

The ado roadmap outlines the planned direction and milestones for the next major versions of the project. This is a living document that will be updated regularly as the project evolves.

🚀 Key Goals

  • Production Ready: Ensure all core operators, actuators and examples can be executed robustly
  • Performant and Human-Centred CLI: A CLI that is responsive and provides helpful feedback on error
  • Agent-Driven Automation: Provide a structured environment for AI agents to seamlessly assist and automate research workflows.
  • Seamless Scaling: Make it easy to scale from single-person working on their laptop, to a distributed team on executing on remote infrastructure.
  • Community & Ecosystem Development: Respond to user needs and empower developers to extend ado

Upcoming Releases

September 2026 Version 2.2.0

Version 2.2.0 will be a maintenance release. It will address a variety of issues found in day-to-day use, along with updates to the bundled plugins and documentation. Some highlights will include:

  • Clearer provenance:
  • Store the experiments and actuators an operation used, including minor and patch version differences.
  • Record the Ray job submission id so logs can be found and the job interacted with.
  • CLI cleanup:
  • Delete an operation's datacontainers with the operation
  • Make samplestore measurements and space properties easier to inspect
  • Improved performance:
  • Load and dump YAML with the C implementation when it is available,
  • Remove performance issues when working spaces with large discrete domains (via ado get space etc.). This includes waiting O(10)s for operations to complete.
  • Plugin updates: Fixes and small capability updates across the plugins

October 2026 Version 2.3.0

Version 2.3.0 will focus on increasing the capabilities of operator plugins. We've noticed that many scripts created for results analysis can't be expressed as operators and hence can't be easily distributed for reuse and fall out of the provenance system.

  • Multi-input, multi-resource-type operators: Enable operator functions to take more that a single DiscoverySpace and also to process resources other than a DiscoverySpace e.g. operation, datacontainer
  • Operator Packaging: Enable bundling operators specific to a given experiment
  • See which operators process outputs of e.g. solve_mip
  • Enable operators to flag that they require certain output metrics to work
  • Batch Operator Application: Provide mechanisms to simplify executing operators multiple times

November 2026 Version 2.4.0

Currently ado only supports cartesian product spaces (full-factorial designs). It also only supports scalar dimensions (scalar experiment input types).

Version 2.4.0 will expand the type of configuration spaces that can be explored. This also means richer experiment definitions (more types, input relationships)

  • Constraints: e.g. Allow the value of one property to be conditional on another
  • Allows fractional factorial designs
  • Enables experiments to forbid certain combinations of parameter values
  • Treatments: Allow a space to specify list of explicit points to explore, giving maximum flexibility in experiment design.
  • Full: Specify the exact entities to sample
  • Partial: Set specific value pairs for some dimensions, while allow full cartesian product for others
  • Vector Properties: Allow easily specifying a dimension that expands to a vector space
  • Currently, you would have to specify each dimension separately
  • Property Groups: Allow grouping properties/dimensions in experiments
  • Enables an experiment be passed a nested dictionary of their parameters rather than flat dictionary
  • This will the need to group input parameters after being passed them

📆 Recent Milestones

August 2026: Version 2.1.0

As we use AI agents to drive research via ado more we've noticed we're creating far more spaces, operations etc., including many failed experiments and trials which we don't need to keep. We're also inundated with agent reports and analysis scripts piling up in our filesystems.

In this release we were adding some features to address these issues:

  • New document resource type for storing agent reports, plans etc.
  • Store reports written for operations or spaces so they can be accessed by collaborators
  • Associate research plans and todos with projects/contexts
  • Agent skills for project maintenance
  • Improve ability of Agents to correctly version plugins, manage their life-cycle and identify versioning related issues
  • Improve ability of Agents to recognise and delete superseded resources, failed and test operations, unused spaces etc. functionality

July 2026: Version 2.0.0

In this release we are making a number of breaking changes in order to address known issues we've encountered since 1.0 and provide a stable platform

  • Refactoring of the CLI to make it more intuitive for humans and agents
  • Include ability to get common stats via ado get
  • Simplified ado show subcommand structure
  • Updated naming and harmonized functions
  • Increased performance of stats queries
  • New StandardActuator baseclass
  • reduces amount of custom code
  • enables synchronous/non-ray execution patterns
  • Enhanced plugin versioning and provenance
  • Improve ability of Agents to correctly version plugins, manage their life-cycle and identify versioning related issues
  • Track plugin versions used in resources

💬 How You Can Help

  • Contribute: Submit pull requests for new features, bug fixes, or documentation improvements.
  • Open Issues: Report bugs, request features, or provide feedback.
  • Spread the Word: Share the project with others who could benefit