DecisionNode

DecisionNode provides a structured, queryable context layer for recording and retrieving development decisions across AI coding tools. It addresses the challenge of maintaining a persistent memory of technical choices, such as design rationales or project constraints, which can be accessed by AI assistants during coding sessions. The tool is intended for developers who use AI-powered coding environments and need a way to ensure that past decisions are not lost or forgotten in future work.

Decisions are captured as structured JSON objects, including fields like scope, rationale, and constraints. json file. When a new coding session begins and the AI needs context, it can search these stored decisions using cosine similarity to retrieve the most relevant past choices. The retrieval process is explicit, with the AI calling the MCP tool to search the decision store, rather than passively injecting decisions into prompts.

DecisionNode integrates with multiple AI coding tools, including Claude Code, Cursor, Windsurf, Antigravity, and any client compatible with the Model Context Protocol (MCP). It offers three main interfaces: a command-line interface (CLI) for terminal-based control, an MCP server for AI integration, and a web UI for visualization. The web UI provides a force-directed graph view, a 2D vector space projection, and a searchable list of decisions, allowing users to see relationships and activity in real time. The tool also supports features such as configurable similarity thresholds for search results, global decisions that apply across projects, full audit logging of changes, conflict detection to avoid duplicates or contradictions, and the ability to deprecate or reactivate decisions as needed.

Installation is via npm, and all operations run locally, with the only external dependency being the Gemini embedding API, which supports a free tier. DecisionNode is open source and distributed under the MIT license, ensuring transparency and local control over data. It is designed to be a persistent, shared memory layer for developers working with AI coding assistants.