AI Tools for Vibe Coding and Everyday Work

New AI tools appear every week. Chasing every name is less useful than knowing which kind of tool your task needs. Start with the outcome, choose a category, and then try one service in that category.

This catalogue brings together assistants, coding tools, document apps, design resources, data tools and infrastructure. It is not a leaderboard: a strong writing assistant is not automatically the best place to build a presentation or deploy a website. The right tool is the one that produces an output you can use and check.

This English WexAI edition adapts the author’s Russian catalogue. Product descriptions are starting points, not a claim that WexAI has benchmarked every service. Availability, model choices and pricing change; confirm them on the linked product page before committing money or sensitive data.

How to use this catalogue

Do not read it as a shopping list. Jump to the part that matches your job and shortlist two or three options. Start with the small stack near the end if you do not yet know what you need.

The biggest gains often come from connecting tools: a transcript becomes source material for a project notebook; extracted page data becomes a spreadsheet. One tool’s output should be a clear input for the next.

Before choosing, answer four questions:

1. What must exist at the end: an answer, a file, a working app or a recurring process?
2. Where are the inputs: a website, your documents, a repository or another application?
3. How much access can the tool have, and what needs your approval?
4. How will you check the result and recover if it is wrong?

Think and research

This family includes conversational assistants, search with sources, agents that take several steps, and browsers with AI assistance built into the page.

Chat assistants

Large language model assistants overlap in what they can do, but differences become clearer on longer tasks. Try your own writing, code or document sample instead of choosing from a generic ranking.

• Claude — Anthropic’s assistant for writing, coding and document work, with projects and artifacts for organising longer tasks.
• ChatGPT — OpenAI’s general-purpose assistant, including text, voice, images and agent-style work where available.
• Gemini — Google’s multimodal assistant, with connections to the Google Workspace ecosystem.
• Perplexity — AI-assisted search with linked sources and deeper research options.

Other models worth knowing

These are alternatives to test for a particular task or budget, rather than subscriptions you need to add all at once.

• DeepSeek — A model family with reasoning and mathematics use cases, including openly available model releases.
• Grok — xAI’s assistant, connected to the X ecosystem and its information flow.
• Kimi — Moonshot AI’s assistant for long-context work.
• Qwen — Alibaba’s assistant and model ecosystem for text, code and multimodal tasks.

Check where data is processed and the terms for your specific plan. A model’s country of origin alone does not tell you how a particular hosted or self-hosted deployment handles your files. Keep confidential information out until you have checked the actual service.

Agents that work through a task

A chat returns a response. An agent can take a goal, use tools and keep working through multiple steps. That makes permission boundaries and a review point important.

• Manus — The original catalogue’s Telegram entry point for an agent handling research, analysis and documents. Verify the bot’s identity before sharing data.
• Hermes — Nous Research’s open agent, designed for deployment in an environment you control.
• Claude Cowork — Delegated work with files and reports, rather than only a back-and-forth chat.
• Genspark — An agent-oriented workspace covering research, documents, slides and other multi-step tasks.
• OpenClaw — An open platform for running a persistent agent on your own infrastructure; check the project’s official distribution and permissions before installation.

Multi-step agents can use more tokens than a single chat response. Treat that as a cost of delegation and set limits before a long run.

AI in the browser

These tools work near the page you are reading, useful for long articles and unfamiliar interfaces.

• Comet — Perplexity’s browser for questions and actions around web pages.
• Dia — A browser from the creators of Arc, built around assistance with tabs and browsing context.
• Claude for Chrome — Claude alongside browser content and selected text.
• Sider — A browser sidebar that brings multiple AI assistants into the same working environment.
• Microsoft Edge — A mainstream browser with integrated Copilot features.

An agentic browser may act through signed-in sessions. Keep banking and sensitive administration separate, and review actions that send, purchase, delete or publish.

Write code and build applications

Vibe coding starts with describing what you want and iterating on generated code. A working preview is only an early milestone: testing, permissions and deployment still matter.

Editors and builders

Visual editors and builders make files, controls and previews visible. They are often an easier starting point than a terminal-only workflow.

• Cursor — An AI coding environment for working with a project and delegating changes to an agent.
• Antigravity — Google’s agent-oriented development environment.
• Qoder — A coding environment aimed at understanding larger project context and working through tasks.
• v0 — Vercel’s builder for turning a description into an application or interface and iterating toward deployment.
• Google AI Studio — A place to work with Gemini and build applications around its API.
• Natively — A builder focused on native iOS and Android applications.

Canvases inside assistants

• ChatGPT Canvas — A workspace for editing text and code in context, rather than only sending new messages.
• Gemini Canvas — Google’s canvas-style editing and preview environment.
• Kimi K2 — A model option associated with coding and agentic tasks; check the current release’s context limits.

Coding agents

These can inspect a project, run commands and change files. Depending on the product, access may be through a terminal, desktop app, IDE or cloud environment.

• Claude Code — Anthropic’s coding agent, with project tools and extensibility for development workflows.
• Codex — OpenAI’s coding agent across development environments, for implementing and reviewing code changes.
• OpenCode — An open coding agent with a choice of model providers and interfaces.
• Claude Cowork — A related option for non-coding file, spreadsheet and reporting work.

Begin with one agent and one project. Parallel agents are useful only when you can understand and review their separate changes. Learn the basics in AI Agents: A Practical Guide before giving an agent wider access.

Coordinating several agents

• Conductor — A Mac environment for parallel agents working in isolated Git worktrees.
• Orca — An open environment with branches, diffs and a built-in browser.
• Rabbithole — A canvas-based workspace that places tasks and knowledge around agents.

The hard part is not opening several agents. It is defining the task, deciding when they should stop and checking the result. Use WexAI’s free first agent lesson to practise those habits.

Keeping your code

• GitHub — Repositories, pull requests, issues and automation through Actions.
• GitHub Desktop — A graphical way to work with GitHub and Git without relying on terminal commands.

Version control may feel optional on a small project. It becomes essential when a generated change breaks a working file. Save a known-good state before a large edit.

Work with voice, files and slides

This family turns existing material into something usable: dictation, transcripts, source-grounded answers, presentations and spreadsheets.

Speech to text

Dictating a thought, transcribing an interview and recording a meeting are different jobs. Choose for the situation rather than treating every speech tool as interchangeable.

• MacWhisper — Audio transcription on Mac, with local processing options for recordings and interviews.
• Wispr Flow — Dictation into applications, with AI-assisted cleanup of spoken text. Listed as “WhisperFlow” in the original catalogue.
• Handy — An open local dictation tool: press a shortcut, speak and insert the resulting text.
• Fireflies.ai — Meeting transcription, summaries and follow-up material for supported meeting platforms.

Text to speech

• ElevenLabs — Speech generation, dubbing and voice tools, with API access.
• Fish Audio — Voice generation and cloning tools, with integration options. Use voice cloning only with appropriate permission.

Your own source material

Use these when an answer should be grounded in the files you supply, not just general model knowledge.

• NotebookLM — Work with supported documents, videos and web sources; ask source-grounded questions and create study outputs.
• Notion — Documents, databases and projects with AI features in the workspace.
• Claude for Word — Claude’s Office tooling for document editing and review; check which applications the current listing supports.
• Google Labs — Google’s experimental tools, including creative and multimodal projects.

A useful pattern is meeting transcription followed by a source notebook. Check consent before recording, then review the transcript before using it as a project reference. The meeting-summary workflow makes the review step explicit.

Presentations

Most presentation generators start from a topic or text and produce a first layout. You still need to check the story, sources and final design.

• Gamma — Generates designed presentations and other content formats from a brief.
• Kimi Slides — Slide creation with different levels of research and generation depth.
• ChatGPT for PowerPoint — A route to building and editing presentations in the PowerPoint workflow.
• Claude for PowerPoint — Office-based presentation assistance; verify current compatibility and editability requirements.
• Figma Slides — Collaborative presentation design with reusable components.

Spreadsheets and numbers

The benefit is working inside the actual file: explaining cells, changing assumptions and checking formulas rather than copying numbers into a disconnected chat.

• Claude in Excel — Spreadsheet assistance for analysing and modifying workbooks.
• Rows — Online spreadsheets with an AI analyst and integrations.
• ChatGPT for Google Sheets — A Sheets marketplace add-on; check its publisher and permissions rather than assuming every ChatGPT-labelled add-on is made by OpenAI.
• Gemini in Google Sheets — Assistance with formulas, organisation and data work in Sheets.

For a tightly controlled brand template, a generic presentation generator may be a poor fit. Building a reusable layout once can be more useful than repeatedly regenerating the whole deck.

Design and interfaces

To avoid a generic AI-generated interface, give the builder a visual system: references, type, colour, spacing and reusable components. A design-token file makes those decisions explicit.

Give the agent a visual direction

• designmd.supply — Turns a website reference into a DESIGN.md-style description of its visual system.
• Neuform — A library of downloadable design systems and references.
• getdesign.md — Design-token files for bringing an established visual direction into a project.
• Impeccable — An open design skill for improving AI-generated interfaces.
• Mobbin — Real application screenshots and user-flow references.
• Dribbble — A design gallery for visual exploration and references.

Use references to explain principles, not to copy someone else’s brand. The diagrams in this WexAI edition use our own lightweight visual treatment.

Build the interface

• Claude Design — Use Claude for describing, exploring and building interface ideas; available design features depend on the product and plan.
• Aura — Turns prompts and visual references into interface designs and code-oriented outputs.
• v0 — Also useful here as an interface generator, not only as an app builder.
• Variant — Explore interface directions and variations through a visual workflow.
• 21st.dev — A registry and ecosystem for UI components, templates and AI-assisted development.
• CodePen — A browser sandbox for quick HTML, CSS and JavaScript prototypes.

Reusable interface pieces

Component libraries can save time, but install only what the design needs. Check licensing, accessibility and bundle cost before copying a visual effect.

• Canvas UI — WebGL-driven visual effects for interfaces.
• Originkit — A library of animated interface components.
• Transitions.dev — Interaction patterns for cards, modals, menus and state changes.
• MotionSites — Prompts and references for animated hero sections.
• Border Beam — A light-trail border effect.
• Liquid Metal — A metallic shader effect for interface elements.
• icons0.dev — Icon search and integration tooling.

Interfaces for AI applications

An AI interface has recurring states: waiting, thinking, calling a tool, streaming a result and asking a person to confirm an action.

• AIcss — CSS-based interface elements for AI conversations and tool states.
• Beautiful UI — Primitives for inputs, human approvals, diffs, tables and related AI interface patterns.
• Thinking Orbs — Visual indicators for agent activity.
• Agentation — Annotate page elements and turn feedback into a brief for an agent.

One clear file of design tokens can prevent more inconsistency than another round of vague “make it better” prompts.

Gather and prepare data

This family extracts useful content from pages and files so a model can work with the substance instead of navigation, advertising and markup.

Clean text from pages and files

• Firecrawl — Converts web content into formats such as Markdown and structured data for downstream processing.
• Jina AI Reader — A reader endpoint for turning web pages into model-friendly content.
• Defuddle — Extracts main-page content while removing surrounding clutter.
• MarkItDown — Microsoft’s file-to-Markdown toolkit, with support depending on format and installed extras.
• Crawl4AI — An open web-crawling toolkit aimed at AI workflows.

A one-off page extraction needs a different setup from a recurring crawl across thousands of pages. Start with the smallest tool that handles your real input.

Ready-made collection platforms

• Apify — A marketplace of extraction tools and a platform for running them, with an SDK for custom work.
• Thunderbit — Browser-based extraction into structured destinations such as spreadsheets.
• RSS.app — Feed generation and automation for following supported web sources.

Build your own collector

• Scrapy — An established Python framework for crawling and extracting data.
• Crawlee — Apify’s framework with queues, retries and browser-oriented collection features.
• Scrapling — Extraction tooling with adaptive approaches to changing page structure.
• AutoScraper — Builds extraction rules from examples of the data you want.
• scrcpy — Displays and controls an Android device from a computer; useful for app-only workflows, but not a web scraper itself.
• Browser Use — Browser interaction tooling for AI agents.

Browser compatibility and protected sites

The original catalogue also includes lower-level browser and HTTP tooling:

• curl_cffi — A Python HTTP client with browser-impersonation capabilities.
• curl-impersonate — A curl variant designed to resemble browser network behaviour.
• CloakBrowser — A Chromium-based browser automation project.
• invisible_playwright — A browser automation project associated with a modified Firefox environment.

Their inclusion is not permission to defeat a site’s access controls. Prefer an official API or export, respect access restrictions, and collect only data you are authorised to use.

Infrastructure for a working project

Models, hosting and databases are the less visible part of the stack. They are what turn a local prototype into a service other people can use.

Model access and keys

• OpenRouter — Access to multiple model providers through a shared API layer.
• OpenAI API — Direct access to OpenAI’s available model and media APIs.
• Groq API — Hosted model inference on Groq’s infrastructure; benchmark latency on your own workload.
• Cohere Reranker — Reranking for improving the ordering of retrieved results in search and retrieval-augmented generation.
• BotFather — Telegram’s entry point for creating a bot and obtaining its token.

An API key can authorise spending. Keep it server-side, outside public repositories and client bundles. Use environment variables or an appropriate secret store, with usage limits and a rotation plan.

MCP and automation

Model Context Protocol connects assistants to external tools and data. Think of it as a connection standard, not a guarantee that every connected tool is safe.

• Zapier MCP — Connects supported application actions to an assistant through MCP.
• Context7 — Brings current library documentation into a coding assistant’s context.
• MCP Market — A directory of MCP servers and integrations.
• Leapter — A visual approach to business logic and tools that agents can call.
• GoHighLevel — A CRM and marketing platform with automation and AI features.
• Is It Agent Ready — A way to assess how accessible a site is to AI agents.

Review permissions before connecting accounts. A tool that can read a document and a tool that can email your customers should not have the same approval policy.

Where to deploy

Choose by workload. A static site, a persistent bot and GPU-heavy processing do not need the same hosting setup.

• Vercel — Application deployment, including frontend and Next.js workflows with previews.
• Railway — A deployment platform for applications, services and databases.
• Netlify — Web deployment with static hosting and supporting platform features.
• GitHub Pages — Static-site hosting from a repository.
• Modal — Cloud compute for Python workloads, including GPU use cases.

Small projects may fit within a provider’s entry-level allowance, but do not assume that every workload is free. Check usage limits, idle behaviour and what triggers charges. This edition uses direct links rather than the original referral links.

Where to store data

• Supabase — A Postgres-based development platform with authentication, storage and other application services.
• Neon — Serverless Postgres with database branching workflows.
• Pinecone — A vector database for semantic retrieval and related applications.

A small stack to start with

Opening the whole catalogue at once is a good way to make no decision. Start with four roles, and add another only when a real task requires it:

• One assistant: Claude, ChatGPT or Gemini for thinking, writing and explaining.
• One source workspace: NotebookLM for working from your own material.
• One research tool: Perplexity when finding and following sources is the job.
• One capture tool: Handy, Wispr Flow or a transcription tool suited to your device and recording needs.

These are alternatives within a role, not a requirement to subscribe to all of them.

When you want to build something of your own:

You do not need fifteen subscriptions. Use a small real task to decide whether the next paid tool removes a bottleneck you actually have.

Quick reference by task

Everyday work

Coding

Infrastructure

Design and data

Before you commit to a tool

Free plans, preview features and privacy settings deserve a closer look than a product screenshot. Check whether uploaded data is retained or used for training, which permissions an integration requests, and whether the plan permits your intended commercial use.

Treat experimental agent workflows as drafts until you have observed how they fail. Keep a human approval step before public communication, spending or irreversible changes. Do not confuse the ability to automate something with permission to do it.

Pricing and product names will change. Some standalone tools will become features inside larger assistants; others will disappear. The durable skill is knowing which family your task belongs to, how to test a candidate and how to replace it without rebuilding your whole workflow.