Skip to main content
Qomero
VIDiscuss your problem
All work

Case study · AI

One state for people and agents.

Qomero's open-source tool. Phases, tasks, decisions and acceptance criteria live as Markdown in the repo, compiled into one shared state.

Status: LIVEApache-2.0

Public Praukron README image: frame-praukron-dashboard.
Public README example; synthetic demo data.

Context

Qomero's open-source tool, Apache-2.0, at github.com/qomero/praukron.

Problem

People and AI agents on one project each read a different state, and finished work blurs into verified work.

Published scope

How Qomero runs its own projects and its clients'.

What we did

Phases, tasks, decisions and acceptance criteria are Markdown in the repository. Compilation is deterministic: same input, same state. Evidence and verification level are recorded separately; superseded decisions are kept.

Verified result

Every Qomero project, yours included, is run through Praukron.

Limits

It's a project-delivery tool, not a solution to a business problem.

Before the build

People and AI agents on one project need the same state: which phase, what is in flight, what has been decided.

Finished and verified are different things. They need recording apart.

Done and verifiedDone and verified. Done; Verified; Acceptance evidence; Verification level.DoneVerifiedAcceptanceevidenceVerificationlevel

Why off-the-shelf failed

Where Praukron came from, and why issue trackers and agent memory tools fell short, is in the README at github.com/qomero/praukron.

What we decided

  1. State lives in the repository. Phases, tasks, decisions and acceptance criteria are Markdown files kept with the code.

  2. Deterministic compilation. No model, no network: same input, same state.

  3. Done is kept apart from verified. Acceptance evidence and verification level are recorded separately.

  4. Old decisions stay. A superseded decision stays on record.

  5. Open source. Apache-2.0, at github.com/qomero/praukron.

Records to stateRecords to state. Markdown; Phases; Tasks; Decisions; Acceptance criteria; Deterministic compilation; Project state; People; AI agents.MarkdownPhasesTasksDecisionsAcceptancecriteriaDeterministic compilationProject statePeopleAI agents
Old decisions stayOld decisions stay. Earlier decision; Superseding decision; Retained.EarlierdecisionSupersedingdecisionRetained

Architecture

Praukron

Phases, tasks, decisions and acceptance criteria as Markdown files beside the code, compiled into one project state.

What the system handles

  • An agent receives a context packet for its task instead of the whole project.
  • People read the project state on a dashboard.
  • Every Qomero project runs through Praukron.
Praukron dashboard, README example: one task with its dependencies, three acceptance criteria (not run, pass, fail) and the blocker that holds it.
Public README example; synthetic demo data.
Praukron dashboard, README example: the Work tab with ready, in progress, waiting for a person, waiting on dependencies and blocked columns, and the root blockers.
Public README example; synthetic demo data.

Result

  • Public under Apache-2.0.

Related solutions AI Copilot

More case studies

  • Screenshot of K-LionBook's Today screen, customer data blurred

    KOVAA

    Sales on spreadsheets, paper and Zalo.

    The same kind of sales and debt system, for your industry.

    Read the case study
  • Screenshot of vietfarmasia.com, running on the ARVUMA codebase

    ARVUMA

    Buyers scattered across channels; no single order record.

    Every channel's orders in one place.

    Read the case study

Sound familiar?

Tell us how that work runs today. We start there.

Discuss your problemAll work