I forked Gemini CLI because it has the best PTY implementation I've ever seen.
Let's be real. Gemini CLI especially with 2.5 sucked for coding. But the execution harness itself - the way it handles interactive terminals, sudo prompts, ssh sessions, TUI apps - that was gold. With Gemini 3 rolling in, I saw what was possible.
So I rewired it to build TerminAI. Multi-LLM support (Gemini, ChatGPT OAuth, LM Studio). Ripped out all Google telemetry. Built a new brain to do anything a human wants to with a computer.
What I ended up with:
- Local-first. Zero telemetry by default. Your prompts stay on your machine
- True PTY. No hanging on interactive commands. Enhanced capability to execute across OS platforms
- A/B/C Approval Ladder. Destructive actions require your explicit approval. You define what's safe
- Multi-platform. Windows parity. PowerShell operates as well as bash
---------------------------------------------------------------------------------------------Some honest notes:
- Some models (like GPT-5.2) hold back. It's their training - they dont know that terminAI is there to protect. Gemini, grok-code-fast-1, and even GLM just let go.
- What IS solid: The core runtime. Brain, safety, multi-LLM, cross-platform, recipes, Linux and Windows.
- What IS not solid yet: desktop app, voice mode, self-hosted cloud relay, MCPs/ extensions (code exists; need to test and finalize)
---------------------------------------------------------------------------------------------Install & Run:
npm install -g @terminai/cli
terminai
In the app:
/llm --to set the llm provider
/llm reset -- to change llm provider. currently requires app restart to take effect
"Audit my docker containers, find the one with high memory, check its logs, summarize errors."
"Fix my drivers...."
"Plan me a detailed itinerary and output as a PDF with a map for..."
or you can always do the good old
"Cleanup my downloads folder"
Happy to go deep on internals.