"AI tool" now describes almost anything with a model somewhere in the stack. A chat assistant is an AI tool. So is a file of open model weights, a Discord server, and a weekly email digest. Those four things have different buyers, different price tags, and different failure modes.
This guide walks through the 19 types of AI tools we use to organize ToolPotion, a directory of more than 13,000 tools, updated daily. For each type: what it actually is, who it's for, and a named example or two where that helps. The 19 sort into four families: things you use, things you build with, things you learn from, and places you go.
The family question is the one to settle first. Evaluating a framework when you needed an app wastes weeks.
The 19 types of AI tools at a glance
| Family | Types | You are trying to |
|---|---|---|
| Things you use | Apps, Mobile Apps, Agents, Platforms | get work done |
| Things you build with | Models, Frameworks, GitHub Repos, Hugging Face, LLMs | ship software |
| Things you learn from | YouTube, Podcasts, Newsletters, Influencers, Books, Learning, Case Studies, Trend Reports | stay current |
| Places you go | Conferences, Communities | meet people who've done it |
Most bad AI purchases are category errors: someone bought a platform when they needed an app, or hired a team to integrate a model when an agent would have done.
Things you use: apps, agents, and platforms
This family is finished software. You sign up, you work, you get output. No engineering required.
AI Apps: the biggest shelf by far
An AI app is a complete product with a model at its core: you bring a task, it brings a result. This is where most people should start, and it's the largest type in the directory. You can browse the full directory of 13,050 AI apps by category, job, or task.
The range is wide. Descript edits video and audio by editing the transcript (delete a sentence of text and the cut happens for you), though for heavy multitrack or color work you'll still reach for a traditional editor. Suno turns a text prompt or your own lyrics into a complete song with vocals in under a minute. Read the license terms before you use the output commercially. Gamma builds presentations and one-page sites from an outline with no design skills required. And ChatGPT remains the general-purpose default for writing, coding help, and image generation, with the standing caveat that a dedicated app usually beats a generalist at any single job.
AI Mobile Apps: phone-first tools
AI mobile apps (2,917 of them at last count) are built around what a phone has that a laptop doesn't: a camera, a microphone, and your location. Think photo editors, voice note transcribers, language practice, and scanning tools. They suit people whose work happens away from a desk: field teams, students, anyone on calls all day.
One check before installing: some "mobile apps" are thin wrappers around a website, with features trailing the desktop version. The reviews usually tell you.
AI Agents: software that does the steps, not just the answer
An AI agent plans and executes multi-step work (reading files, calling APIs, clicking through interfaces) instead of returning a single response. The directory lists 553 AI agents, and it's the fastest-moving type of AI tool right now.
Two examples from opposite ends. Claude Code is a coding agent that reads a codebase, edits files, and runs commands from the terminal. It assumes you're comfortable on a command line, and you should still review its diffs like any pull request. n8n lets technical teams wire AI into business workflows, mixing code with a visual builder. If "technical teams" doesn't describe yours, a simpler automation app like Zapier with its 9,000+ app connectors is the gentler on-ramp.
Agents need supervision in proportion to what they can touch. Start them on low-stakes work.
AI Platforms: the layer a company standardizes on
An AI platform is infrastructure: model access, permissions, deployment, and governance in one place, on which a company builds many things. The 94 AI platforms in the directory are bought by IT and engineering leaders, rarely by individuals.
The practical difference from an app: an app solves one job today; a platform is a commitment. Platform choices are sticky, so this is the one type where a slow, boring evaluation is the right call.
Things you build with: models, frameworks, and open source
Everything in this family assumes there's an engineer in the room.
AI Models: the engine, not the car
An AI model is the trained system itself, usually consumed through an API and priced per use. The directory tracks 324 AI models, from frontier general-purpose systems to single-task specialists. Claude Opus is a hybrid reasoning model aimed at serious coding and agent workloads, with a 1M-token context window. FLUX.2, from Black Forest Labs, generates images with unusual control over composition and style. Model pages matter even if you never call an API: the model behind an app determines much of what the app can do.
AI Frameworks: scaffolding for builders
AI frameworks (128 listed) are the libraries developers use to connect models to real products: orchestration, retrieval, evaluation, agent loops. Strictly a developer purchase. The honest caveat for the whole category: frameworks churn fast, so pin your versions and expect breaking changes.
AI GitHub Repos: open source worth your stars
The 169 AI GitHub repos in the directory are open-source projects (agent toolkits, scrapers, self-hosted interfaces) that you can run for free. The price you actually pay is maintenance: you become your own vendor, security patches included.
AI Hugging Face: open weights you can own
The Hugging Face listings (71 tools) cover open models you can download, fine-tune, and self-host. This route suits teams with hard privacy requirements or unusual domains that off-the-shelf APIs handle poorly. Budget for GPUs and for the ops work: self-hosting a model is running production infrastructure, not installing an app.
LLMs: a small shelf of specific checkpoints
The LLMs type is the smallest in the directory (four entries) and deliberately niche: specific community language models and fine-tunes rather than the big-name systems, which live under AI Models. A representative example is FFMPerative-7B, a Llama 2 fine-tune that drives video-editing operations like cropping and resizing from natural-language commands. Interesting if you need exactly that. Skip the shelf entirely if you don't.
Things you learn from: keeping up without drowning
Eight of the 19 types of AI tools aren't software at all — they're how practitioners stay current. The trick is rationing them.
For watching and listening: the 254 AI YouTube channels are the best way to see a tool demonstrated before you commit to a trial, and the 166 AI podcasts fit the commute: interviews and analysis rather than demos. The 98 AI newsletters are the efficient default for most people. Most of them overlap heavily, so subscribe to two and unsubscribe without guilt. The 133 AI influencers (practitioners posting on X) are the fastest signal and the worst noise ratio of the bunch. Treat their claims as leads to verify, not conclusions.
For reading deeper: the 79 AI books cover foundations that outlast any release cycle, and the 104 AI learning resources (courses and structured tutorials) are the route when you need a skill rather than an opinion. Two smaller types earn their place at work. The 81 AI case studies document how real companies deployed AI and what broke, which makes them the rare honest genre in AI content. The 108 AI trend reports supply the charts and adoption numbers you need when a strategy deck or budget request is due.
Places you go: conferences and communities
The last two types are about people, not products.
The 65 AI conferences listed are where announcements land and where you can compare vendors side by side in an afternoon. They're expensive in both money and time, so choose by who attends, not by who's on stage: the hallway conversations are the product.
The 59 AI communities (Discord servers, forums, practitioner Slack groups) are the opposite trade: free, always on, and the fastest place to get an implementation question answered by someone who hit the same wall last month. Quality varies widely between communities. Lurk for a week before you judge one.
How to choose your starting point
- If you have a specific job to get done (edit a video, draft outreach, build a deck), start with apps, filtered by that task.
- If the job is a repetitive multi-step process, look at agents next, and give one a low-stakes trial run.
- If you're building a product, your shopping list is models plus frameworks, with open-source repos and Hugging Face weights as the self-hosted alternative.
- Whatever you pick, add one newsletter and one community so next quarter's decision is easier than this one's.
The directory's browse pages exist for exactly this: every type page above filters further by category, job role, and task, so you can go from "what kinds of AI tools exist" to a shortlist in a few clicks.
Frequently asked questions
What is the difference between an AI app and an AI agent?
An app responds to you: you act, it answers, you act again. An agent carries out a multi-step task on its own (planning steps, using other software, and checking results), with you supervising rather than driving. Agents are more capable and more dangerous in equal measure, which is why they warrant review checkpoints that apps don't need.
What is the difference between an AI model and an AI platform?
A model is a single engine: one trained system, usually accessed through an API and priced per use. A platform is the environment a company runs many models and applications in, with access control, deployment, and governance attached. Individuals buy model access; organizations buy platforms.
How many types of AI tools are there?
There is no standard taxonomy. ToolPotion uses 19 types across four families: tools you use, tools you build with, resources you learn from, and places you go. The exact count matters less than the family: knowing whether you need finished software, developer components, or a learning resource eliminates most of the directory before you start comparing.
Which type of AI tool should a beginner start with?
Apps, without much doubt. They need no setup, most have free tiers, and abandoning one costs nothing. Pair that with one newsletter to stay oriented. Leave agents until you can name a repetitive process worth automating, and leave models, frameworks, and platforms to whoever ships your software.
