I use Claude a lot, but when Claude Cowork came out my first thought was:
Why am I paying $20/month for an AI that automates my own computer?
If an AI is clicking buttons, reading files, running shell commands, and controlling apps on my machine… why does that need to live in the cloud?
So I built EverFern.
It’s basically a free, open-source alternative to Claude Cowork — an AI agent that lives on your desktop, sees your screen, controls apps, automates browser tasks, and runs workflows.
Except:
Runs locally (supports Ollama / LM Studio)
No subscription
MIT licensed
Your files stay on your machine
What it actually does
You can even connect to any cloud provider if you want(We support even free providers like Nvidia NIM, etc)
🖥️ Computer use
Takes screenshots, finds buttons, clicks, types, navigates apps — basically uses your computer like a human would. You just describe the task in plain English.
🧠 Skills system
Reusable community-made workflows. Think plugins, except they're simple shareable markdown files you can make yourself.
🌐 Browser agent
Navigate websites, scrape data, fill forms, automate repetitive browser stuff without writing automation scripts.
🐧 Linux VM sandbox
Shell commands run in an isolated environment so the agent can’t accidentally nuke your system.
🤖 Multi-agent debate
For harder tasks, multiple agents argue over the best approach before execution. Sounds gimmicky, but it catches surprisingly dumb plans.
⚡ Workflow builder
Chain actions together, save them, schedule them.
Stack (if you care)
Electron + Next.js frontend, LangGraph for orchestration, MCP integrations.
Works with local models (Ollama / LM Studio) or cloud providers (OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter).
Current state
Windows installer is live.
macOS installer isn’t ready yet, but you can run it with:
npm install && npm run dev
Still early and rough around the edges, but the core computer-use + browser automation actually works surprisingly well.
I mostly built this because I didn’t want to keep paying for something that felt like it should run locally anyway.
GitHub: Here's the Repo, Hope all like it
Would genuinely love feedback — especially on the skills system because that’s the part I’m still figuring out.
Top comments (1)
"I didn't want to pay $20/mo for something that should run locally" is the exact frustration driving a lot of the current AI-tooling backlash - flat monthly subscriptions for AI features feel wrong when the underlying compute is bursty and per-use. You don't run the agent every minute of every day, so paying a fixed rent for occasional use is a bad deal, and building the OSS local version is the rational response. The bet you're making is real: own the runtime, pay only for the model calls you actually make, no rent.
I'm in the same camp on pricing philosophy, just landed on a different point: usage-based instead of subscription. The thing I build, Moonshift, is a multi-agent pipeline that takes a prompt to a deployed SaaS on your own GitHub + Vercel, and the model routing means a full build lands ~$3 flat per run instead of a monthly seat - you pay for the build, not for the calendar. Same "don't pay rent for bursty work" instinct as your local-first approach, different tradeoff (managed orchestration + deploy vs run-it-yourself). First run's free, no card, if you ever want to compare. Genuine question since you went local: how are you handling model costs - BYO API key, or local models? The local-model path is where "free" actually holds up vs just shifting the bill to your own API key.