Project Context & Change Intelligence for AI Coding

Coding agents reconstruct different pictures of the same project. Shared memory and retrieval alone are commoditizing.

Teams re-explain intent, agents pull the wrong context, and changes can drift from architecture, decisions, and specs without anyone noticing.

Polterware keeps a project model of declared intent and observed software, with evidence and provenance, then turns that model into usable agent context and change checks.

From the model, Polterware compiles the smallest useful context for the task at hand and serves it to any coding agent, so every agent works from the same grounded picture. When changes disagree with architecture, decisions, or specs, drift checks surface the mismatch, and an audit trail shows what context led there.

Context compilation, agent access, drift checks, and run audit are what we are building, grounded in a project model people can inspect. Those are product direction, not finished claims.

Polterware is open source. We are not shipping a paid product yet. We are building in the open so the model, the context path, and the drift checks can be inspected, forked, and improved.

The work centers on consistent context across coding agents, evidence-backed provenance for what goes into that context, and change verification against architecture, decisions, and specs, including an audit trail from context delivered to change to evidence.

We are deepening the project model of declared intent and observed software, with evidence people can inspect.

Next we are building evidence-backed project facts, task-scoped context for agents, drift checks, and a clear audit of what each agent saw.

Alongside that come stronger change gates, deeper spec-driven workflows, and cloud surfaces for teams. Architecture views may remain a way for people to inspect the model. They are not the category.