AI assistantsCAN I VIBECODE
Claude?
BUILD FIRST
- A non-sensitive or read-only version of the core experience
- A manual review step wherever correctness or trust matters
- A narrow companion tool that removes a real piece of work
LEAVE FOR LATER
- Regulated decisions, payments, or irreversible actions
- Claims of correctness, compliance, or uptime that the MVP cannot support
- Sensitive integrations until security, support, and auditability are ready
RESEARCH NOTES
What this tool actually does.
Claude is an AI assistant developed by Anthropic, available as a web app (claude.ai), desktop app, and iOS/Android apps. It is a conversational assistant that writes and edits content, generates and debugs code, analyzes uploaded files (PDF, Word, Excel, images), searches the web for current information, and executes code in a sandbox. Signature features include Artifacts (shareable, interactive documents, visualizations, and mini-apps generated from chat), Projects (persistent-context folders that organize chats and documents), voice mode, and research artifacts with citations in styles such as APA, MLA, and Chicago. It integrates with Google Workspace (Gmail, Calendar, Docs), JIRA, Zapier, Intercom, and other tools via MCP connectors. Claude Code is a terminal-based coding agent; Cowork lets Claude operate on local files and cloud apps with per-step user approval. The platform API expo
RESEARCHED BUILD PROMPT
Give this to your coding agent.
Build a self-hosted conversational AI assistant web app inspired by the core workflow of Claude by Anthropic: a persistent chat where users delegate writing, coding, data analysis, and research to an LLM that can produce interactive artifacts and search the web.
Workflow: a user starts a chat, optionally attaches files (PDF, DOCX, XLSX, images) or selects a Project (a named folder with persistent context and shared instructions); the assistant streams responses, can call a web-search tool and cite current sources, can render code or data as an interactive artifact (chart, mini-app, document) in a side panel, and can run sandboxed code for analysis. Users can switch between voice input and typing and share artifacts via public links.
User-facing requirements: streaming chat UI with markdown and code highlighting; file upload with text extraction; artifact panel rendering HTML/JS or charts; per-response citation list; Projects list with persistent history; voice input via the browser speech API; shareable artifact links; settings for model choice and API key.
Technical requirements: Node.js/TypeScript backend exposing a chat API that proxies to an LLM provider (Anthropic or OpenAI-compatible) with tool-calling for web search and sandboxed code execution (Docker or a restricted subprocess with timeouts and no network access); SQLite or Postgres for messages, projects, and artifacts; React frontend with SSE streaming; a search API for web lookup.
Implementation approach: use the provider's streaming API with function/tool calling; store conversations as JSONB; render artifacts in a sandboxed iframe; use a job queue for long-running analysis.
Do not copy the Claude brand, logo, proprietary UI, or any Anthropic assets; build an original UI and name. Out of scope: multi-user auth/teams, native mobile apps, an MCP connector marketplace, enterprise admin, model training/fine-tuning, and computer-use control of user machines.
Include a README covering setup (env vars, provider key, Docker for the sandbox), architecture overview, and important limitations (sandbox security, token costs, web-search reliability, no offline model).