ProjectFlow AI

Turn human domain knowledge into approved, evidence-backed plans before Codex builds.

Comment

ProjectFlow AI

ProjectFlow AI is an open-source, AI-native project management and software-delivery platform being developed by Aurentium Technologies Ltd.

It helps teams capture human domain knowledge, turn that knowledge into evidence-backed requirements, approve the exact plan, and then allow Codex to work only within a controlled and reviewable execution cycle.

The product preserves the full chain:

Problem → questions → answers → evidence → requirements → approved plan → Agile delivery → authorised Codex execution → testing → human review → release

The problem

AI coding agents can build software quickly, but they can also begin against incomplete requirements, unsupported assumptions, stale approvals, or unclear authority. Traditional project-management tools track tasks, but usually do not preserve why a requirement exists, who supplied the knowledge, what evidence supports it, or exactly what an AI agent was authorised to change.

The solution

ProjectFlow AI acts as the control and collaboration layer between human project knowledge and AI software execution. Human stakeholders and AI can propose questions, domain experts provide evidence, the team reviews generated requirements, and required parties approve an immutable plan version before Codex begins work.

Codex then operates through a restricted execution cycle with explicit repository, branch, path, tool, network, test, cost and checkpoint boundaries. The system records what Codex did, why it stopped, what tests passed, and what human review is required before work continues or a release is recorded.

Planned prototype capabilities

  • Secure organisation, project, member and guest workflows
  • Human-written and AI-suggested discovery questions
  • Immutable stakeholder answers and evidence fragments
  • Evidence-backed requirements, assumptions, risks and acceptance criteria
  • Versioned project plans and exact approval snapshots
  • Agile backlog and sprint planning
  • Approval-gated Codex execution cycles
  • Restricted repository, file, tool, network and budget scope
  • Human checkpoints, testing and technical/stakeholder review
  • Change control and requirement-to-release traceability
  • Open-source, self-hostable deployment

Initial demonstration

The first demonstration uses a two-person project involving a software developer and a chiropractor acting as a non-technical domain expert. It uses only generic professional and business knowledge, with no patient-identifiable health information.

Development status

The production-oriented planning dossier is complete and the full working prototype is currently being implemented with Codex and GPT-5.6. This description will be updated to reflect the exact working features, testing results and final demonstration once the build is complete.

Built With

Share this project:

Updates