Artifact

Artifact is an open source workspace where researchers pair with AI to understand papers and other reading material. As you read, a background agent automatically builds and maintains your personal journal — a living record of the concepts, definitions, and connections you're picking up along the way.

Load an arXiv paper, a web article, or your own PDF into a reader beside the assistant, and ask questions grounded in the full text. Your journal grows quietly in the background as you read and chat, and every future session is grounded in what it already knows about you — no re-explaining where you left off.

Bring your own keys. Configure providers in Settings and use the models you already pay for—Anthropic, OpenAI, xAI, or any OpenAI-compatible inference API (OpenRouter, Fireworks, Together, Sail, etc.). One selector lists every model you add.

Data stays on your machine: reviews, annotations, uploaded PDFs, and graph state live in your browser's IndexedDB—there is no server-side database, no accounts, and no telemetry.

Try it now: withartifact.com — no install, just add your API keys and start reading.

Run locally

Prefer to run it yourself (for development or self-hosting)? Clone the repo and:

npm install
npm run dev

Open localhost:3000 and add your API keys under Settings.

Using Artifact

Open anything. Artifact reads from multiple sources so you can work with whatever you have in front of you:

• arXiv ID — paste an ID like 2401.12345 and the paper loads straight from arXiv.
• Paper or web URL — drop in any paper link or arbitrary web page; the content is cleaned and rendered alongside the assistant.
• Local PDF upload — pick a PDF from your machine and it's stored locally in your browser (IndexedDB), ready to read and query.

Once a source is open, you get the content next to Assistant, Notes, and Explore. Ask selection-scoped questions, and run Analyze once per source to fill prerequisites, related candidates, and the shared graph. Use Discovery to browse how analyzed papers relate and what to open next.

Ambient Journal. As you read and chat, Artifact quietly builds a personal wiki of what you're learning. The Journal captures key concepts, definitions, and connections across sessions, then feeds that context back into future chats so the assistant remembers what you already know — without you having to re-explain or re-link anything.

Contributing

Contributions are welcome. Open an issue before large changes so the approach can be discussed.

The chat agent is a ReAct-style loop over a tool registry in src/tools/. New or improved tools (arxiv-search, web-search, rank-results, save-to-graph) extend what the model can do without forking the core loop.

npm run lint     # ESLint
npm run test     # Vitest
npm run build    # Type-check + production build

Tech stack

• Framework — Next.js 16 (App Router, Turbopack), React 19, TypeScript
• PDF — react-pdf / pdfjs-dist (full text extraction and selection)
• Web pages — @mozilla/readability + DOMPurify for cleaned, safe rendering
• Markdown — react-markdown, remark-gfm, remark-math, rehype-katex
• Graph — React Flow (@xyflow/react) with custom paper nodes and relationship edges
• Styling — Tailwind CSS 4, shadcn/ui
• Storage — IndexedDB via Dexie (client-side, per-browser; PDFs stored as blobs)
• AI — Anthropic, OpenAI, xAI, OpenAI-compatible APIs (streaming chat + structured generation)
• Paper search — Semantic Scholar (primary), arXiv API (fallback)

Project structure

src/
├── app/
│   ├── api/
│   │   ├── arxiv-metadata/    # Fetch paper metadata from arXiv
│   │   ├── arxiv-search/      # Search arXiv for related papers
│   │   ├── chat/              # Streaming agentic chat (multi-provider, ReAct loop)
│   │   ├── generate/          # Structured generation for analysis pipeline
│   │   ├── models/            # Available model catalog
│   │   ├── pdf/               # PDF proxy (CORS) + validation
│   │   ├── web-content/       # Fetch and clean arbitrary web pages
│   │   └── wiki-schema/       # Serve and edit the ambient wiki schema
│   ├── discovery/             # Knowledge graph page
│   ├── journal/               # Ambient wiki / journal view
│   ├── review/[id]/           # Reader (PDF or web page + right panel)
│   └── settings/              # API key management
├── components/
│   ├── related-works-graph    # Interactive graph (React Flow, custom nodes + edges)
│   ├── graph-canvas           # React Flow canvas with paper nodes and relationship edges
│   ├── chat-panel             # Chat with streaming + analysis progress
│   ├── prerequisites-panel    # Prerequisite checklist + study guides
│   ├── right-panel            # Tabbed panel (Assistant / Notes / Explore)
│   ├── pdf-viewer             # PDF renderer with text selection + annotations
│   ├── web-viewer             # Readability-cleaned web page reader
│   ├── annotation-list        # Annotation management
│   └── sidebar                # Navigation + review history
├── hooks/
│   ├── use-auto-analysis      # Analysis trigger + status tracking
│   └── use-explore-data       # Reactive access to graph/prerequisite data
├── tools/
│   ├── arxiv-search           # Paper search tool (Semantic Scholar + arXiv)
│   ├── web-search             # General web search tool
│   ├── rank-results           # Result ranking tool
│   └── save-to-graph          # Knowledge graph persistence tool
└── lib/
    ├── explore-analysis       # Multi-phase analysis pipeline
    ├── explore                # Graph types, storage, merge logic
    ├── reviews                # Review sessions + chat message persistence
    ├── annotations            # Annotation CRUD
    ├── deep-dives             # Advanced learning sessions
    ├── wiki                   # Ambient knowledge base: types, storage, ingest, journal agent
    ├── models                 # Model + provider definitions
    ├── client/db              # Dexie (IndexedDB) schema and queries
    └── client/pdf-blobs       # Local PDF blob storage (IndexedDB)

License

MIT