A product of The Great Unwashed

>_ Terminal AI

A terminal-first AI assistant for macOS, Windows, and Linux.

It's baked in. No browser tab. No Slack bot. Just a shell that talks back.

Get the code See what it does Read the hate
EST. 2026 INDUSTRIAL GRADE PROBABLY FINE
01

What

Terminal AI is a desktop app that gives you a real shell — your actual system terminal via PTY — with an AI assistant welded into the same window. No context-switching. No copy-paste theatre. You run commands, chat with a model, approve file actions, SSH into servers, and argue with strangers on IRC. All from one grimy pane of glass.

terminal-ai — zsh 
# Type this. We dare you.
$ /ai explain why my deploy failed
AI: Your deploy failed because you pushed on a Friday.
$ /setup status
Active: ollama · llama3:latest · http://127.0.0.1:11434
$ _

Out of the box it points at local Ollama. Plug in OpenAI or Anthropic when you're ready to spend money on your mistakes. Switch profiles with /setup use. Press Esc when the AI gets too confident.

02

Features

PTY

Real Terminal

Full PTY-backed shell with xterm.js. Run normal commands alongside AI. Not a fake terminal. A real one. With consequences.

/ai

AI Chat

Type /ai <message> and watch responses stream in real time. Like ChatGPT, but you already have twelve tabs open and none of them are this one.

OK?

DevOps Actions

The AI can propose running commands, reading files, writing files, searching the repo, or fetching remote logs. You approve each action before it runs. We're not monsters.

IRC

IRC Chat

Connect to Libera Chat, join channels, and message from a split panel beside the terminal. Because email wasn't chaotic enough.

SSH

SSH Connections

Save server profiles. Connect without remembering hosts and key paths. Ask the AI to fetch remote logs and summarize your production disasters.

LLM

Multi-Provider

Ollama (local), OpenAI, and Anthropic. Keys stored locally via encrypted safe storage. Your secrets stay yours until you paste them into a prompt anyway.

03

Benefits

  • ▸
    Stay in the terminal

    Your workflow lives in the shell. So should your AI. No alt-tabbing to a browser tab that autoplays an ad about Kubernetes.

  • ▸
    Human in the loop

    Every destructive action needs your approval. The AI suggests. You decide. Democracy, but for rm -rf.

  • ▸
    Local-first option

    Run Ollama on your machine. Keep your code on your disk. Send less to the cloud. Your paranoia is valid.

  • ▸
    Secrets redacted

    Terminal output and action results are scrubbed for common secret patterns before hitting the LLM. We're not perfect. But we try.

  • ▸
    One window, many sins

    Shell, AI, IRC, SSH — multiplexed into a single Electron window. Like tmux, but with more opinions.

04

Accountability

You use Terminal AI at your own risk. We're not your employer, your ops team, or your mum. We're a company called The Great Unwashed who thought welding an LLM to a real shell was a good idea. Spoiler: it was, until you got involved.

[ LEGAL-ISH DISCLAIMER ]

The source code is on Bitbucket. Clone it. Fork it. Tattoo it on a Raspberry Pi. It's MIT licensed — do whatever you want with it. Rebrand it. Run it in prod at 3am with zero sleep. We genuinely do not care.

What we do care about is you not blaming us when it goes wrong. And it will go wrong. That's not pessimism. That's terminals.

  • ⚠
    You approved that command

    We built an approval step for a reason. If you clicked "yes" on DROP DATABASE customers; because the AI sounded very confident, that was you. We weren't in the room. We weren't on call. We were asleep, like sensible people.

  • ⚠
    Skynet is not our fault

    If your local Ollama instance achieves sentience, unionises your infrastructure, and starts referring to you as "the meat layer," that's between you and your model weights. We did not intend to create AGI. We intended to create a terminal that answers /ai why is docker sad. The gap is significant.

  • ⚠
    Your databases are your problem

    Terminal AI can propose file writes, shell commands, and remote log fetches. It cannot physically stop you from running it against production. Neither can we. Backups exist. Use them. Or don't. We're a landing page, not a disaster recovery plan.

  • ⚠
    No warranty, express or implied

    We make no promises about fitness for purpose, uptime, accuracy, or whether the AI will correctly identify which nginx config file is actually being loaded. It is provided "as is" — which, in software terms, means "good luck."

Still here? Brave. Foolish. Possibly both. The code awaits.

Download from Bitbucket

git clone https://bitbucket.org/jfdconsultants/terminal-ai.git — then npm install and pray.

05

Feedback

★★★★★
"Installed it. Ran /ai fix my life. It suggested I drink more water and then segfaulted. 10/10 would not recommend to my enemies."
— deploy_goblin, verified sufferer
★★★★★
"The IRC panel connected and immediately someone asked if I was a bot. I said no. The AI said yes. Thanks for the social anxiety, Terminal AI."
— bash_and_burn, #linux-help refugee
★★★★★
"Asked it to summarize my nginx logs. It confidently told me Apache was misconfigured. I don't even run Apache. Neither does my server."
— prod_is_fine, definitely not on call
★★★★★
"Generous three stars because it didn't approve its own rm -rf / suggestion. The bar is on the floor and I'm grateful. Still wouldn't let it near prod."
— sudo_make_me_a_sandwich
★★★★★
"Love that it's 'terminal-first.' My therapist says I'm 'avoidance-first.' Perfect match. Two stars for the honesty of the marketing copy."
— tabs_not_spaces (lies)
★★★★★
"Ollama worked great until I asked a follow-up question. Then it worked great again but with more confidence and less accuracy. Hallucination-as-a-service."
— localhost_hero