Everfern

Everfern: Install and first job

This is page 2 of 8 for Everfern. It installs the app on Windows 11 or a Mac, turns on the safety switches before Everfern touches your screen, connects a model, and walks you through your first safe job.

Checked 2026-11-14 against Everfern v1.3.0, published 22 September 2026.

Install on Windows 11

Step 1. Download the installer. Open the official v1.3.0 release page:

https://github.com/Everfern-AI/Everfern/releases/tag/v1.3.0

Scroll to Assets and click EverFern.Setup.1.3.0.exe, which is about 327 MB. Ignore the .dmg, .zip, .AppImage, .deb, .rpm, .blockmap, and .yml files, because those are for other computers or for the app's updater.

What you should see a file named EverFern.Setup.1.3.0.exe in your Downloads folder.

Step 2. Run the installer. Double-click the file. Do not right-click it and pick Run as administrator.

This guide could not confirm whether the installer is signed by a known publisher. If Windows shows a blue box that says Windows protected your PC, it means Windows does not know who made the file, so check that you downloaded it from the GitHub page above. If you still want to go ahead, click More info, then Run anyway, but if you are not sure, stop here.

This guide could not confirm what the installer's screens look like, so follow them and keep the default choices.

What you should see the installer finishes, and Everfern shows up in your Start menu.

Step 3. Open Everfern. Open the Start menu, type EverFern, and open it.

Windows does not ask a separate permission question before an app moves the mouse or types. Once Everfern is open and you give it a job, it can control your screen.

What you should see an Everfern window opens. This guide could not confirm what the first screen shows, and it may ask you to pick a model first. If it does, read "Turn on the safety switches first" before you continue.

Install on a Mac

Step 1. Find out which Mac you have. Click the Apple menu in the top-left corner, then About This Mac. If you see a line that says Chip with "Apple M" in it, you have an Apple silicon Mac, and if you see Processor with "Intel" in it, you have an Intel Mac.

What you should see either the word Apple (an M chip) or the word Intel.

Step 2. Download the right file. Open the official v1.3.0 release page:

https://github.com/Everfern-AI/Everfern/releases/tag/v1.3.0

Scroll to Assets.

There are also .zip versions of the Mac app, but use the .dmg.

What you should see a file whose name starts with EverFern-1.3.0 and ends in .dmg in your Downloads folder.

Step 3. Put it in Applications. Double-click the .dmg. If the window shows the EverFern icon and an Applications folder, drag the icon onto the Applications folder. This guide could not confirm the exact look of that window.

What you should see EverFern in your Applications folder.

Step 4. Open it for the first time. Open Applications and double-click EverFern.

This guide could not confirm whether the Mac app is approved (notarized) by Apple. If your Mac says it cannot check the app or that it is from an unidentified developer, click Done or Cancel. Then open System Settings, go to Privacy & Security, scroll down, and click Open Anyway next to the EverFern message, but do that only if you downloaded the file from the GitHub page above.

What you should see an Everfern window opens.

Step 5. Screen permissions (you can wait on these). A Mac will not let any app see or click your screen until you allow it, so to use computer use, Everfern will likely need two switches in System Settings, under Privacy & Security:

This guide could not confirm which of these Everfern asks for, or when. If it asks before you have done the safety switches below, you can say no for now and turn each one on later in those same two places.

What you should see Everfern open, and you know where those two switches are. Turn them on only after the next section.

Turn on the safety switches first

Do all of these before you let Everfern control the screen. A sentence in chat like "don't change anything" is a request, not a lock. The research behind this list is in the safety checklist.

1. Close personal apps and log out. Computer use looks at your real screen and can click anything on it. Close your email, chat apps, banking, shopping, password manager, and personal documents, and log out of email, your bank, and stores in every web browser on this computer. A web page can hide instructions aimed at an AI agent, so if the agent can reach a logged-in mailbox, that page has a way to ask for it.

What you should see only Everfern and an empty desktop. No email, bank, or store is logged in anywhere.

2. Make one folder for the agent. Make a new folder with nothing personal in it.

Tell Everfern in every job to work only in that folder. This guide did not find an Everfern setting that locks it to one folder, and computer use can open anything you can open, so the folder is a habit, not a wall.

What you should see an empty everfern-tryout folder.

3. Keep the approvals. The README says Everfern "pauses for your confirmation before anything destructive," and that new tools it writes for itself wait for "a human-approval step before anything activates." This guide could not find a documented switch for these approvals, so there is nothing to turn on. In the v1.3.0 code, only commands on your real computer always show a card. Other pauses depend on the model choosing to ask, so keep your shortcut ready. If you ever see a setting that turns approvals off or lets it run without asking, leave it off.

When it asks, read the whole question. Say no if it mentions deleting, moving, sending, buying, logging in, your home folder, Documents, email, or a new tool.

What you should see no setting changed to "no approval" or "run without asking."

4. Set a spend cap. A local model (Ollama or LM Studio) has no bill, and a free Google Gemini key needs no card. If you use a paid key, set a monthly cap at that provider before you paste the key into Everfern:

This guide did not find a dollar cap inside Everfern. Do not sign up for Everfern's Pro or Max plans for this tryout.

What you should see a saved monthly limit at your provider, or no paid key at all.

5. Know how to stop it. While Everfern works, the send button in the chat box turns into a square stop button. If the agent is using the mouse, quit the app the normal way instead. Keyboard shortcuts work best, because the agent may be using the mouse.

This matters. An August 2026 release note says it fixed a "computer use win-key loop," which means the agent had been getting stuck pressing the Windows key over and over.

What you should see you have tried quitting Everfern once with your shortcut and opened it again, before any job.

Mac only: now you can turn on the Accessibility and Screen Recording switches from "Install on a Mac."

Connect a model

The README says you can "run fully offline with Ollama or LM Studio, or connect to OpenAI, Anthropic, DeepSeek, Gemini, OpenRouter, Nvidia NIM, Mistral, Groq, and more," and that you can switch providers anytime. It also says your keys are encrypted and kept on your computer in ~/.everfern/store. The v1.3.0 setup screen names ~/.everfern/config.json instead.

Everfern has a settings page, which the README and a July 2026 release note both mention, and in v1.3.0 the place to choose a model provider is the Models & Providers section of settings. In v1.3.0, the setup screens that can appear the first time you open Everfern start with Choose your engine, where AI Provider is for a cloud key and Local Engine is for Ollama or LM Studio.

Computer use works by taking pictures of your screen, and Everfern's website shows a screenshot() step before each click. So keep two things in mind:

Use a cloud key, or a model on your own computer from the next section.

A cloud key. The easiest free start is a Google Gemini key from Google AI Studio, as in the main guide. The README lists Gemini, but this guide did not test it with Everfern, and Google may use free-tier prompts to improve its products. For a paid OpenAI or Anthropic key, set the spend cap first (safety switch 4). Paste the key into Everfern's settings, and never paste it into the chat.

What you should see your provider chosen in Everfern's settings, and the key not in any chat message.

Or use a local model

Install Ollama and download a model first, using Run a local model with Ollama. Skip Everfern's Install Ollama Automatically button, because this guide could not confirm what that button downloads. Computer use needs pictures. The sizing notes are in Local models by hardware, and on about 16 GB the tag to pull is mimo-v2.6:9b. A model whose name ends in :cloud is not on your computer.

In the v1.3.0 setup screens:

  1. On Choose your engine, click Local Engine.
  2. On Select AI Provider, click Ollama.
  3. Leave the Server URL (optional) box empty, and Everfern uses Ollama's local address, http://localhost:11434. Click Continue.
  4. On Local AI Models & Hardware, open Installed Models, click your model, and click the Continue with button that shows its name.

This guide could not confirm the setup screens after that. To change the model later, use Models & Providers in settings.

What you should see Provider Status says Connected above the word Ollama, with your model selected. If it says Offline, start Ollama and click Retry.

LM Studio instead. Download LM Studio from lmstudio.ai/download, get one model, and start its local server, whose address is http://localhost:1234/v1. Then pick LM Studio instead of Ollama in step 2.

What you should see LM Studio chosen in Everfern, with your loaded model selected.

Small local models are weak at long jobs, as the hardware page explains. A slow or confused agent on 8 or 16 GB is normal and does not mean you set it up wrong. The same Ollama address works for every agent in this guide, as Connect it to your agent explains.

Your first safe job

Start with two tiny jobs that touch nothing personal, and keep your hand near your quit shortcut.

Job 1: chat only. Type:

Without using any tools or controlling my computer, tell me in three sentences what you can do.

What you should see a short answer and no mouse movement.

Job 2: one small click job. On Windows, type:

Open Notepad. Type "hello from Everfern". Then stop. Do not save the file. Do not open any other app or website.

On a Mac, use the same words but say TextEdit and "a new document" instead of Notepad.

Watch the whole time, since Everfern's README says it shows its reasoning live. If it opens anything you did not ask for, quit the app with your shortcut.

What you should see Notepad or TextEdit opens, the words appear, and the agent stops. Close the note without saving.

After that: one file in your folder. If the chat shows Select folder as project..., choose everfern-tryout first. Page 3 explains why this matters. Then try:

Make a text file named first-note.txt in my everfern-tryout folder. Put three tips for drinking more water in it. Do not touch any other folder.

Then open the everfern-tryout folder yourself. If the file is not there, it may have been written inside Everfern's Linux VM, and this guide could not confirm where those files end up. The next page explains what Everfern's code says about where its files go.

What you should see only first-note.txt is new, and nothing else on your computer changed.

Sources

Opened 2026-11-14.

Next

Continue to Files, the browser, and memory.

Checked 14 November 2026 · prices, limits, and versions change