frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Why I forked Gemini CLI - a FOSS Cowork alt that *is* the OS

https://github.com/Prof-Harita/terminaI
1•ProfHarita•1h ago
Hi HN, I'm the maintainer of TerminAI — a local-first AI that operates your computer, not just chats about it.

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.

Comments

ProfHarita•1h ago
OP here. Three things that are hard to convey in the title:

1. This isn't a "run my prompt" tool. TerminAI maintains a persistent, live "System Spec" of your machine—your installed binaries, your paths, your shell quirks. When standard CLIs fail, it pivots and writes custom Python/JS on-the-fly to solve the problem. It doesn't just call tools; it builds the tool it needs.

2. Every action is auditable. Every thought, plan, tool call, and outcome is logged to a local JSONL file you control. When something goes wrong, you have a forensic trace of why it made that decision. That's the trust guarantee—not "just believe me," but "here's the receipt."

3. Why this has to be open source. %% And why I began developing this in December %% The CLI democratized compute. The GUI democratized access. We're now at the third shift: the OS is becoming an intelligent partner. But who owns that partner? If the "System Operator" is rented from a closed provider, you're a tenant in your own machine. TerminAI is the open-source answer. It's the bash of the Agentic Age.

Happy to go deep on the Brain architecture or the Zero-Trust Cloud Relay.

Ask HN: How to find a sponsor?

1•TySchultz•1m ago•0 comments

San Francisco coyote swims to Alcatraz

https://www.sfgate.com/local/article/san-francisco-coyote-alcatraz-21302218.php
1•kaycebasques•1m ago•0 comments

People cannot "just pay attention" to (boring, routine) things

https://utcc.utoronto.ca/~cks/space/blog/tech/PeopleCannotPayAttention
1•todsacerdoti•2m ago•0 comments

All your OpenCodes belong to us

https://johncodes.com/archive/2026/01-18-all-your-opencodes/
1•jpmcb•2m ago•0 comments

The Code-Only Agent

https://rijnard.com/blog/the-code-only-agent
2•emersonmacro•4m ago•0 comments

Bring Back Ops Pride

https://charitydotwtf.substack.com/p/bring-back-ops-pride
1•mooreds•4m ago•0 comments

Show HN: Knowhere – Rust-Based SQL Engine with TUI and GUI (Built on DataFusion)

https://saivarunk.github.io/knowhere/
1•waxsum8•8m ago•0 comments

With this tool, you can enjoy NAS functionality even without a NAS

https://quicksend.chat/
1•foodhome•10m ago•0 comments

HN should be dark mode by default

1•busters4•11m ago•2 comments

Show HN: All Paul Graham essays as a single ePub for e-readers

https://tomyanz.com/pgessaysepub/
2•tomyanlol123•16m ago•0 comments

We grew an XR conference to 11,000 attendees. Here's why we walked away

https://1upsummit.com/blog/built-largest-xr-conference-part-of-problem/
1•makercameron•19m ago•1 comments

Offensive Windows IPC Internals 3: ALPC (Advanced Local Procedure Call)

https://csandker.io/2022/05/24/Offensive-Windows-IPC-3-ALPC.html
2•aragonite•21m ago•0 comments

Trzsz-SSH (tssh): An SSH Client Alternative in Go

https://github.com/trzsz/trzsz-ssh
1•csmantle•31m ago•0 comments

The "Bucket Bumping" problem of airline tickets, and how to minimise your fare

https://www.dodgycoder.net/2026/01/the-bucket-bumping-problem-of-airline-tickets.html
2•damian2000•33m ago•0 comments

Pre-Agent Nostalgia

https://www.pcloadletter.dev/blog/pre-coding-agent-nostalgia/
3•ronbenton•34m ago•1 comments

Privacy Engineering at Scale: Building Automated Data Retention Systems

https://medium.com/@sandhyavinjam/privacy-engineering-at-scale-building-automated-data-retention-...
1•sandhyavinjam•45m ago•0 comments

There's no corpus large enough

https://www.swiftcraft.io/articles/no-corpus-large-enough
1•rad_val•47m ago•0 comments

Mermaid as a programming language for AI agents

https://twitter.com/xiaoxxchan/status/2011825791408226618
2•xxchan22•50m ago•0 comments

Show HN: N.codes – Let users generated mini-apps in your app

1•yungookim•53m ago•0 comments

Show HN: PixelMotion:AI video generation with Sora 2, Veo 3.1, and 9 more models

https://www.pixelmotion.io/
1•anotherbuilder•56m ago•0 comments

Tesla investigates whether its self-driving technology caused traffic violations

https://www.latimes.com/business/story/2026-01-16/tesla-is-probing-if-its-self-driving-technology...
2•MilnerRoute•56m ago•0 comments

I built a proper UI for e-com product image generation (chat doesn't scale)

https://aimockupgenerator.co/
1•dairis•56m ago•1 comments

Ask HN: Could we replace Job Descriptions with actual Git Issues?

1•A1aM0•57m ago•0 comments

Gemini Research MCP Server

https://github.com/fortaine/gemini-research-mcp
2•gfortaine•1h ago•0 comments

New Framework Proposes Spacetime as an Emergent Irreversible Information Process

https://zenodo.org/records/17911993
1•emsti•1h ago•2 comments

Show HN: Webship.dev – custom web systems, legacy code and performance

https://webship.dev/
1•pragidi•1h ago•0 comments

Disqus but Using GitHub. Cool Idea

https://giscus.app/
2•razodactyl•1h ago•0 comments

High-speed train crash in southern Spain leaves at least 21 dead

https://www.theguardian.com/world/2026/jan/18/high-speed-train-crash-in-adamuz-cordoba-southern-s...
7•perihelions•1h ago•1 comments

Show HN: Shared Device Group – scheduler-level GPU sharing for Kubernetes

https://github.com/sceneryback/shared-device-group
1•ChuangLabs•1h ago•0 comments

Clip to Siglip > Migrating Our 200M+ MultiModal Embeddings

https://huggingface.co/vulturelabs/vector-rosetta-clip-vit-base-patch32-to-siglip-vit-base-patch1...
1•teocalin37•1h ago•1 comments