CPO IDE — development of an environment where people, LLMs and AI agents work together on projects. We are building a knowledge hierarchy, storage of decisions and causal relationships, a change history and time-based snapshots of project state. The work includes model API adapters, context preparation from project materials, agent task management and resource accounting. An Electron and React desktop shell and individual knowledge, history and execution components have been implemented; integration and testing continue. CPO IDE continues our work on development tools but is not a new version of Miracle.
Projects
Development environment · LLMs and AI agents
CPO IDE. Projects, knowledge and AI agents.
An environment for people, LLMs and AI agents to work together: tasks, knowledge, decision history and verification remain part of the project when models or work sessions change.
In development · free release planned for 2027
knowledgedecisionscontinuity
01
Challenge
A long-running project needs to preserve more than code: what needs to be done, why a decision was made, what information supported it and what has been verified. When work is shared between people, models and agents, this context repeatedly needs reconstructing. CPO IDE is being developed to preserve project continuity throughout that work.
02
Our contribution
NPF I.V.A. is designing and developing its own environment: a project state model, knowledge and decision storage, task orchestration, context preparation for LLMs and a desktop interface. Its central architectural principle is that project state exists independently of a particular AI provider. Changing models should preserve active tasks, decision rationale and work history. This continues our experience in building development tools; CPO is not a new version of Miracle.
03
Knowledge and decision relationships
Knowledge is designed as connected objects: requirements, plans, decisions, materials and work results. Their sources, versions, scope of applicability and relationships matter. We are researching and implementing knowledge hierarchies and causal links to trace a requirement through to a decision, understand why it was made and identify conclusions that need revisiting when conditions change.
04
Context and LLMs
We are developing API adapters for LLMs and different mechanisms for agent task execution. Context is built for a specific task from identified versions of project materials. File access, context inclusion and disclosure to an external model are handled separately. This supports control over the knowledge being shared and work with different models while retaining project memory outside an individual chat.
05
Agent orchestration
We separate task formulation, action authorisation, execution, verification and acceptance. An agent receives a bounded scope and available resources; its report is linked to observations and checks. We are developing task start, interruption and continuation, resource accounting and execution history. The aim is to let a person understand what was done, on what basis and which results have actually been substantiated.
06
History and recovery
The architecture provides for project state versions, views at particular points in time and change comparisons. History should retain both the sequence and meaning of actions: which decision changed, why and what work it affects. State reconstruction is separated from repeating external actions: reading history again must not rerun operations. This area is being implemented and checked through individual scenarios.
07
What has been built
An Electron and React desktop shell and components for the project model, context construction, execution and event recording have been implemented. The interface includes project knowledge and history views, with isolated demonstration scenarios prepared. Work continues on joining components into user workflows, validating the knowledge hierarchy and testing integrations ahead of public release.
08
Product development
We plan to release a free version of CPO IDE in 2027 and include it in the Russian Software Register. Paid add-ons are also planned; their scope and timing will be defined as the product develops. Development and testing are ongoing.
Project in the timeline
Start with your challenge