Systems & tools
Governed AI memory with CTX Continuity
CTX Continuity addresses a practical problem in working with AI: the conversation ends, but the project, its decisions and the reasons behind them continue. My interest is in memory that can be inspected, corrected and carried between working environments.
What should survive a conversation?
A useful working memory needs to preserve the project’s structure: decisions, constraints, unresolved questions and the source of each claim. Keeping a transcript is a different task. A long transcript can retain every sentence while making the important correction difficult to find.
CTX approaches continuity as a set of explicit records. Their source, status and relationship to other records matter as much as the summary. That makes it possible to ask why a statement is present and whether it still applies.
Corrections belong in the record
Suppose an early discussion names one material for an object and a later decision replaces it. The next working session should use the correction while preserving enough history to explain the change. Repeating both versions as equally current would preserve text and lose the decision.
This example captures the design principle: new corrections change what the system carries forward. A model’s observation, a user’s instruction and an externally sourced fact remain distinguishable.
Portable context, with boundaries
The continuity architecture uses inspectable records, versioned state and selective retrieval. Relevant context can move with the work instead of depending entirely on one conversation window. Visibility and consent remain part of the information’s handling.
This page describes the project’s purpose and governing method. It is an introduction to my systems work, with related examples in SeamWatch’s evidence handling and GEMFORM’s explicit design constraints.