The latest articles on DEV Community by Xia Chen (@chenxiachan). https://dev.to/chenxiachan https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4048034%2F7b28c2c3-94d8-4d6e-aa74-e10bc1081d78.jpg https://dev.to/chenxiachan en Xia Chen Sun, 26 Jul 2026 14:26:56 +0000 https://dev.to/chenxiachan/i-made-llm-context-editable-a-graph-where-the-wires-are-the-prompt-2afl https://dev.to/chenxiachan/i-made-llm-context-editable-a-graph-where-the-wires-are-the-prompt-2afl <p>Most LLM interfaces present a conversation as a transcript.</p> <p>That works until the conversation branches.</p> <p>One branch contains useful research. Another is a discarded hypothesis. A third is an unrelated detour. The interface may show where they came from, but the user often cannot see exactly which branches will be included in the model's next request.</p> <p>I wanted to make that state editable.</p> <h2> Wires are the context </h2> <p>I built an open-source prototype called ThoughtDAG.</p> <p>Every question-and-answer exchange is a node on an infinite canvas. An edge is not merely a visual relationship: it determines which upstream nodes are included when the next request is constructed.</p> <p>When a user asks from a node, ThoughtDAG walks its incoming graph, orders the relevant ancestors, and turns them into the message sequence sent to the selected model.</p> <p>Delete an edge and regenerate the same prompt, and the deleted branch is no longer part of the model's context.</p> <p>[Insert the edge-deletion GIF here]</p> <h2> Why a DAG? </h2> <p>I did not want an autonomous system that silently reorganized the user's ideas.</p> <p>The graph is directed because context flows from earlier reasoning into later reasoning. It is acyclic because a node should not recursively include itself in its own history.</p> <p>This makes three operations explicit:</p> <ul> <li>branching: explore an alternative without modifying the original path;</li> <li>merging: synthesize several selected branches;</li> <li>pruning: remove irrelevant context before regenerating an answer.</li> </ul> <p>The graph is therefore not an agent workflow. It is a visible, human-controlled representation of model memory.</p> <h2> Local-first by default </h2> <p>The canvas is stored in IndexedDB in the browser. There is no account or hosted database.</p> <p>ThoughtDAG can connect to Ollama for local inference, or to an arbitrary OpenAI-compatible endpoint. PDFs remain on the user's machine; only extracted text is sent when a remote model is deliberately used.</p> <p>The project is MIT licensed and built with React, TypeScript, React Flow, Zustand, Express, and the Vercel AI SDK.</p> <p>Repo: <a href="https://github.com/chenxiachan/thoughtdag" rel="noopener noreferrer">https://github.com/chenxiachan/thoughtdag</a></p> <p>The question I am testing now is whether people actually want to control context manually, or whether they would rather delegate memory selection to retrieval and agents.</p> <p>I would be interested in critiques of that interaction model, especially from people building long-running LLM interfaces.</p> ai showdev opensource llm