Lists

Genuinely free — no trial that expires, no card up front. Some are open source, some are free tiers a company funds deliberately. Both count; neither is a fourteen-day countdown.

Weights or code you can read, run and modify. Useful if you need to self-host, avoid vendor lock-in, or answer a compliance question about where your data goes.

Nothing leaves your machine. Slower than the hosted equivalents and limited by your hardware, but the only honest answer when the data can't go to a third party.

Coding agents, completion, review and the infrastructure underneath. The split that matters: tools that edit your code, and tools that serve models to your code.

Tools that write and edit code with you. Roughly three shapes: inline completion, chat in the editor, and agents that take a task and come back with a diff.

Drafting, editing and rewriting. Worth separating the ones that fix errors from the ones that generate text — most writers want the first and get sold the second.

Reading, note-taking, research and writing support. Check your institution's policy before using any of these on assessed work — that's on you, not the tool.

Literature search, citation mapping and grounded question answering. The ones that cite sources are the only ones worth using here.

UI, graphics, decks and asset generation. The generation tools get the attention; the editable-output tools get used twice.

SEO research, content operations, ads and outbound. Priced for teams, mostly — check whether the seat count matches yours before committing.

Generation, editing, captions and avatars. Generation gets the headlines, but captioning and repurposing are what actually save working editors time.

Recording cleanup, transcription, editing and clipping. Fix the audio first — no amount of clever editing rescues a bad room.

Things a team of one to ten can actually afford and run without an IT department. Enterprise platforms are deliberately excluded — they need a procurement process you don't have.

Assistants with a real free tier, or open-weight models you can run yourself. Free hosted tiers come with rate limits; local models come with hardware requirements. Pick your constraint.

Text-to-image models and the products built on them. The real decisions: hosted or self-run, licensed training data or not, and whether you need readable text in the output.

Text-to-speech, cloning and dubbing. Only clone a voice you own or have written permission to use — several of these will terminate your account for it, and in some places it's illegal.