The code survives the session. The reasoning should too.
Disclosure: I’m building Wye. This article was drafted with AI assistance. Have you ever opened a feature you built with an agent and wondered: why did we decide to do it this way? The code is there. The tests pass. But the conversation that explained the odd exception, the rejected alternative, or the customer promise is buried in an old session. When you build with a coding agent, decisions…
When building a feature with an agent, the reasoning behind the decisions often gets lost in old sessions. The code and tests may still be available, but the conversation explaining the odd exceptions, rejected alternatives, or customer promises can be difficult to find. This issue arises because decisions happen throughout the development process, some during conversation and others during implementation. The history of why certain decisions were made can become harder to trace.
The goal of building Wye is to create agentic memory that both people and agents can communicate with, inspect, and evolve together. This memory grows alongside the work and provides a way to capture durable knowledge rather than just preserving every conversation. By installing a vault alongside the project and using skills to teach agents to read and contribute to its memory, the additional context needed for understanding the reasoning behind decisions can be stored.
Wye allows for the preservation of decisions, their reasons, scope, and relationships to existing knowledge. It checks incoming knowledge against its neighbors in the graph to ensure consistency and avoid conflicting instructions. If there are contradictions, they are made explicit for people to review and decide how to handle them.
As decisions are superseded, their history remains available, including the reasons why things previously worked differently. This consistency check happens before future agents inherit any ambiguity.
Wye utilizes Markdown as its canonical storage, allowing product knowledge to live alongside the code. It can be reviewed in a diff, retained in Git, and read or edited without relying on Wye's interface. At runtime, Wye represents the knowledge as a connected graph, enabling retrieval based on relationships between decisions, requirements, constraints, and dependencies. This graph structure supports retrieving relevant information even when it uses different language.
The runtime graph also supports retrieval beyond the visual map view, allowing for more specific queries based on the context of the decision, such as the review requirement and its rationale in the publishing example. The balance Wye aims to strike is capturing enough reasoning to preserve continuity without creating a second documentation job.
The project is early, local-first, and licensed under Apache-2.0. It currently works with Claude Code and Codex, and the repository includes a tour and setup instructions. The author is particularly interested in hearing about real decisions that have been lost between agent sessions.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
