refern. VS ContextForge.dev

refern.'s answer

I'm an artist, and too often I was frustrated: I couldn't find the right pictures, I struggled to organize and keep track of my reference images, and I was juggling too many apps when all I wanted to do was draw. I wanted one tool that could do it all, with every component well integrated. So I made refern.

ContextForge.dev's answer:

ContextForge was born from a simple frustration: AI coding agents forget everything the moment a session ends. Every new conversation meant re-explaining the same architecture, naming conventions, and past decisions. ContextForge was built to give AI agents a permanent, searchable memory through the Model Context Protocol — so knowledge is captured once and reused forever, across sessions and projects. It even dogfoods its own memory to help build itself.

refern.'s answer

The flow of collecting references for art usually goes like this:

find references and inspiration online -> save and organize those images -> use them on their own or in a moodboard -> reuse them for the next project or share them with others

Normally that means a different app for each step. refern does all of it in one place, with every part built to work together.

ContextForge.dev's answer:

ContextForge is memory that lives at the MCP layer, so it works across every AI coding agent at once — Claude Code, Cursor, GitHub Copilot, ChatGPT, and Windsurf — not just one. Save a decision once and any client recalls it later with semantic search. It goes beyond a note store: automatic git sync turns your commits and PRs into searchable knowledge, plus task tracking, snapshots, and team sharing — all through a single MCP server you add with one command.

refern.'s answer

Most tools solve only one part of the reference workflow, so artists end up juggling several. refern combines them in one app: Eagle-style organization and search, a PureRef-style infinite canvas, and an Obsidian-style graph of how your references connect. It never copies your files, so it doesn't double your disk usage. It runs on Windows, macOS, and Linux, works fully offline, and is a $30 one-time purchase with no subscription. Built in Rust, it stays fast even on libraries with hundreds of thousands of images.

ContextForge.dev's answer:

Most memory tools are tied to a single agent or are just a key-value store. ContextForge is MCP-native, so it's portable across all your AI tools; it adds git sync so your codebase history becomes searchable context automatically; and it includes team features (shared spaces, collaborators) that solo-memory tools lack. Setup is one command, there's a genuine free-forever tier with no credit card, and paid plans start at just $9/month.

refern.'s answer

Individual artists and creative professionals: illustrators, concept artists, digital painters, designers, and photographers. They collect and organize large libraries of visual references, build moodboards, and study from reference. refern is built for solo creators rather than teams, spanning hobbyists and students through working professionals.

ContextForge.dev's answer:

Software developers and engineering teams who use AI coding assistants — Claude Code, Cursor, GitHub Copilot, ChatGPT, Windsurf — and are tired of re-explaining their project, architecture, and conventions every session. It fits solo developers working across multiple projects as well as small teams that need shared, persistent context.

refern.'s answer

refern is a Tauri v2 desktop app with a Rust backend and a React 19 (with React Compiler) plus TypeScript frontend. Data is stored locally in SQLite (rusqlite) with FTS5 full-text search. The interface uses Tailwind CSS, Zustand, and TanStack Query, with motion/react for animation. The infinite canvas and graph view are built on react-three-fiber and three.js. Thumbnailing, perceptual hashing, and color and visual similarity run in Rust with rayon and SIMD.

ContextForge.dev's answer:

Next.js 16 (App Router), React and Tailwind CSS for the dashboard, hosted on Vercel. Supabase (PostgreSQL) with pgvector powers the semantic vector search, and Deno edge functions serve the API. Embeddings use OpenAI text-embedding-3-small. The MCP client is a Node.js package (contextforge-mcp) on npm, implementing the Model Context Protocol.