Detailed Analysis
A developer operating under the handle tjqscott has released a lightweight convention tool called Dory, designed to address one of the most common friction points in AI-assisted software development: the loss of project context when switching between AI models or starting new chat sessions. The tool consists of a single markdown file placed in a repository's root directory, which maintains an append-only changelog of project decisions and progress. At the start of any new session, the developer instructs the AI model to read the file, immediately restoring working context. At the end of a session, the developer prompts the model to append a summary line describing what was accomplished.
The problem Dory addresses is structurally inherent to how large language model chat interfaces currently operate. Context windows are finite and stateless between sessions, meaning every new conversation begins from zero. Developers working on complex, long-running projects face a recurring tax on both their time and token usage as they re-establish the state of their codebase, explain architectural decisions already made, and reconstruct the reasoning behind earlier choices. The creator specifically cites credit consumption and cross-model switching — moving between Claude and Gemini to manage costs — as the direct motivation, highlighting how the economics of using multiple AI providers introduces its own form of cognitive overhead.
Dory's design philosophy reflects a broader pattern of developers building minimal, convention-based solutions rather than relying on platform features or dedicated tooling. By storing context as a plain text file tracked by Git, it becomes version-controlled, portable, and model-agnostic. The file travels with the codebase rather than living in any vendor's proprietary system, which gives it durability independent of any single AI provider's product decisions. This positions it as infrastructure rather than a product integration.
The tool connects to a wider trend in the AI developer tooling ecosystem where practitioners are increasingly building personal systems to manage the gap between how LLMs are designed and how they are actually used in sustained, professional workflows. Features like memory, persistent context, and project awareness have become central selling points for AI coding assistants such as GitHub Copilot Workspace, Cursor, and Claude's own Projects feature — all of which attempt to solve the same underlying problem at the platform level. Dory represents the grassroots counterpart to those commercial solutions: no vendor lock-in, no subscription tier, no setup, just a file.
The emergence of tools like Dory signals that despite significant investment by AI companies in context management features, meaningful gaps remain in real-world developer workflows, particularly for users who work across multiple models or who prioritize cost control over convenience. The fact that a single markdown file can provide meaningful quality-of-life improvement in daily AI interactions suggests the current generation of AI interfaces still places substantial burden on the user to maintain continuity — a burden that developers are solving themselves with remarkably low-tech interventions.
Read original article →