Free software worth your time, judged honestly
What you will find here
- Sixteen free tools worth your time, each with its real drawback named
- Plain explanations of what each one actually does, and who it is for
- Why nothing on this list has paid to be here, and never will
This is the software we think is worth your time, judged the same way as everything else on this site. Free and open wherever possible, running on machines you already own, and every single entry names its real drawback.
That last rule is the important one. If we cannot tell you the catch, we do not understand the thing well enough to recommend it, so it does not get written up.
Nobody pays to be on this list. There are no sponsors, no affiliate links and no arrangements of any kind. If you use every tool here we earn exactly nothing, and the nothing is the point.
Almost everything below is free and open, meaning anyone can read how it works and run it themselves. One paid tool is mentioned because it has earned its place in daily use. We are paid nothing for saying so.
Open source means the instructions the program is built from are published for anyone to read, use and change. It usually also means free.
Self-hosted means it runs on a computer you own, in your house, rather than on a company's computers somewhere else. Your information stays where you can see it.
The one we would point at first
Buzz
A shared workspace, a bit like a team chat and project tool, except it is built so that both people and automated helpers are treated as proper members with their own verified identity. You can run the whole thing on your own machines.
Good for: keeping a permanent, tamper-evident record of who did what. Every message and every action is signed and written into a list that can be checked afterwards and cannot be quietly edited. Because you host it yourself, you own the record. It was released by Block, Jack Dorsey's company, in July 2026 and given away openly, aimed squarely at the closed tools most teams are locked into.
The catch: it is early, and its makers say so plainly. It is roughly at version 0.4, the phone apps are unfinished, notifications are not done, and setting it up yourself is genuine technical work rather than an installer you click. It is not a finished product yet.
Find it at github.com/block/buzz
And to answer the question everybody asks: it is not cryptocurrency. The signing is there so the record can be checked, nothing more.
Running AI on your own computer
Ollama
The easiest way to start running AI language models on your own machine. You fetch a model with one command and then talk to it.
Good for: getting started without fuss. It handles the downloading and managing for you, and it presents itself in a standard way that almost every other tool knows how to connect to. It runs on fairly modest computers.
The catch: it is a friendly layer over a more powerful engine underneath, so you trade some fine control for the convenience. Its default settings are not always the best ones for your particular machine, and a large model still needs real memory that no amount of convenience can conjure up.
Find it at ollama.com
llama.cpp
The engine that made running these models on ordinary hardware practical in the first place. Most of the friendlier tools, Ollama included, are sitting on top of it.
Good for: efficiency and control across an enormous range of machines, with or without a dedicated graphics card, and detailed control over how much memory a model uses.
The catch: it is lower level. You handle the model files and the build settings yourself. More power and more setup, which is exactly why most people meet it through something friendlier.
Find it at github.com/ggml-org/llama.cpp
Open WebUI
A web page for talking to the models running on your own computer, which looks and feels like the commercial chat assistants.
Good for: making a model on your own machine feel like a real product. It handles several users, keeps separate histories, lets you switch models and can answer questions about documents you give it. All of it stays on your hardware.
The catch: it is the face, not the brain. You still need a model running behind it. It also gains features steadily, which means more to set up and more to keep updated than a simple chat box.
Find it at github.com/open-webui/open-webui
ComfyUI
A tool for making images and video on your own graphics card. Instead of one button, you connect up the steps yourself on a canvas.
Good for: complete control, and being able to save and share the exact recipe that produced a result so it can be repeated.
The catch: connecting the steps yourself is a genuine learning curve. There is also a real safety warning: add-ons and shared recipes can run instructions on your computer, so only use ones from people you trust. It also wants a capable graphics card to be enjoyable.
Find it at github.com/comfyanonymous/ComfyUI
Voice and cameras, kept at home
Home Assistant voice control, with Whisper and Piper
Voice control for your house that runs entirely on your own hardware. One piece turns your speech into words, the other turns the reply back into speech.
Good for: private, offline voice commands. No account, and no recording of your voice ever leaves the building.
The catch: recognising speech on your own machine is harder work and less forgiving than the big commercial speakers. How well it hears you, and how quickly it answers, depends on your hardware, and setting it up is more involved than plugging in a smart speaker.
Find it at home-assistant.io
Frigate
Records your security cameras on your own machine and works out what it is looking at, so it can tell you about a person or a car instead of every branch that moves.
Good for: useful alerts with no monthly camera subscription and no footage leaving the house. It fits neatly alongside Home Assistant.
The catch: it really wants its own dedicated setup, and realistically a small extra piece of hardware to do the recognising smoothly. Getting the cameras and detection areas configured is real work rather than plug in and go.
Find it at frigate.video
Tools that write code and run tasks
An agent is an AI assistant that has been given permission to actually do things, rather than only talk about them. Run a command, edit a file, open a web page. The list of things it is allowed to do is the part that matters, and it is also the part that carries the risk.
Aider
An AI helper that works alongside you in a terminal window, editing the files in a real project and recording each change as it goes.
Good for: quick, well aimed edits across a real project, with every change recorded separately so you can review or undo it. It works with models running on your own machine or hosted elsewhere.
The catch: it is only as good as the model behind it, and it will confidently make wrong changes, so you are the reviewer every single time. The best results come from paid hosted models, and models running on your own machine are noticeably weaker at this particular job.
Find it at aider.chat
OpenHands
A more independent version of the same idea. It will write code, run commands and drive a web browser to finish a whole task rather than one step.
Good for: handing over a complete job instead of the next line, and you can run it on your own machines.
The catch: independence cuts both ways. It runs real commands and can wander off course, so you keep it walled off and you watch it. Results vary a lot depending on the model behind it.
Find it at github.com/All-Hands-AI/OpenHands
Dify
A workshop for building your own AI tools, using a visual builder rather than writing everything from scratch.
Good for: getting from an idea to something that works quickly, without connecting every piece by hand, and it runs on your own machines so your data stays yours.
The catch: the convenience costs you some fine control, and running the whole thing yourself is heavier than a single small program. You can outgrow the way it does things once your needs get specific.
Find it at dify.ai
LangGraph
A toolkit for programmers building AI processes with several steps that need to remember where they are and branch depending on what happens.
Good for: reliable multi-step processes you can reason about and test. The right choice once a single simple loop is no longer enough.
The catch: it is a toolkit for people who write code, not a visual builder, so there is a learning curve. It also brings a larger family of software along with it, which has its own weight and changes often.
Find it at github.com/langchain-ai/langgraph
n8n
A tool for wiring different services together so things happen automatically. Think of it as a visual switchboard, with AI steps built in as first class parts.
Good for: connecting apps and services into automatic sequences using a visual editor, hosted on your own machine so your passwords and data stay there. It knows how to talk to a very large number of services.
The catch: despite appearances, it is not fully open source. The code is readable but the licence restricts what you may do commercially. That is fine for running it yourself at home, but read the terms before building a business on it. Large automations also get hard to maintain.
Find it at n8n.io
Small hardware that helps AI run at home
NVIDIA Jetson Orin Nano
A small board with a proper graphics processor built in, for running vision and smaller language models directly on a device rather than sending anything away.
Good for: noticeably more AI capability than a Raspberry Pi, which makes it a good fit for robotics and camera projects that need to think for themselves.
The catch: it costs considerably more than a Raspberry Pi, it runs NVIDIA's own version of Linux which is its own thing to learn, and it draws more power than most people imagine when they hear that something runs "on a small device".
Find it at nvidia.com
Raspberry Pi AI HAT+
An add-on board that clips onto a Raspberry Pi 5 and gives it a dedicated chip for recognising things quickly.
Good for: cheap, low power AI help on hardware you may already own. A natural partner for home camera work of the kind Frigate does.
The catch: it only speeds up models that have been specially prepared for it, so you work within a supported list rather than running anything you like. It is a helper for specific jobs, not a general purpose graphics card.
Find it at raspberrypi.com
Coral USB Accelerator
A small stick that plugs into a USB port and takes the recognising work off a Raspberry Pi or a small computer.
Good for: very cheap, very low power help with spotting things in camera images. The classic partner for a home camera recorder on modest hardware.
The catch: it is narrow. Models must be specially prepared for it, so it suits a small family of camera related jobs and does nothing at all for language models. Its supporting software has not seen much attention lately.
Find it at coral.ai
Still to come
More categories are on the way: gadgets, radio equipment and computer hardware. Same rules, and an honest drawback named on every single one.
This page in the original Kyber Cypher voice: TECH WATCH