Master Decision Node with our step-by-step tutorial, detailed feature walkthrough, and expert tips.
Explore the key features that make Decision Node powerful for developer workflows.
DecisionNode stores development decisions as structured JSON rather than as loose markdown notes. The website example includes fields such as id, scope, decision, rationale, and constraints, which makes each record easier for AI tools to interpret later. It then embeds the decision text as a vector and stores it locally in vectors.json. This makes the decision searchable by meaning, not just by exact keywords.
The website explicitly lists compatibility with Claude Code, Cursor, Windsurf, Antigravity, and any MCP client. Because it exposes an MCP server, the retrieval workflow depends on the AI coding client calling the search_decisions tool. The site also notes that retrieval is explicit: decisions are not automatically injected into the system prompt. This is useful for developers who want searchable memory without permanently bloating every prompt.
Yes. The website schema lists DecisionNode as accessible for free, with an offer price of 0 USD, and identifies the license as MIT. The install URL points to the npm package decisionnode, and the homepage shows installation via npm i -g decisionnode. Users should still account for the Gemini embedding dependency: Google lists gemini-embedding-001 as free of charge on the Gemini API free tier, $0.15 per 1M input tokens on paid Standard, and $0.075 per 1M input tokens on paid Batch.
DecisionNode converts decision text into embeddings using Gemini’s gemini-embedding-001 model, then compares future queries against stored vectors using cosine similarity. The website states that embeddings are 3072-dimensional and that the web UI can project them into a 2D vector space using UMAP. It also supports a configurable minimum similarity threshold to filter out irrelevant results. The example workflow on the site shows a prior UI decision returned with a 94% similarity score.
DecisionNode includes conflict detection before adding a new decision. According to the website, existing decisions are checked at 75% similarity so near-duplicates and contradictions can be caught. This does not replace human review, because the team still needs to decide whether two records truly conflict. It does, however, create a useful guardrail when many decisions accumulate across multiple AI coding sessions.
Now that you know how to use Decision Node, it's time to put this knowledge into practice.
Sign up and follow the tutorial steps
Check pros, cons, and user feedback
See how it stacks against alternatives
Follow our tutorial and master this powerful developer tool in minutes.
Tutorial updated March 2026