Everfern: Pick it and plan

This is the first of eight pages for Everfern, and it helps you decide whether Everfern fits you before you download anything. It covers who Everfern suits, what to have ready, which models it can use and what they cost, and the catches you should know about first.

Note: Everfern is a very small project. On 14 November 2026 it had about 101 stars and 12 forks on GitHub, while the main guide's pick, Agent Zero, has far more users checking it. Everfern's README says shell commands run inside an isolated Linux VM and that it pauses before anything destructive. In the v1.3.0 code, only commands run on your real computer always stop for your approval. Every other pause, including before deleting files or using your screen, happens only if the model decides to ask. Computer use is the big feature, and it drives your real desktop: it clicks and types on your actual screen, in your actual apps, not inside that VM.

This guide is for you if you picked Everfern instead of Agent Zero. Everfern (Everfern-AI/Everfern) is a free desktop app under the MIT license. You describe a job in plain English, and it plans the steps and then carries them out by looking at your screen, moving the mouse, and typing. It also has a built-in browser agent called Navis.

Checked 2026-11-14. The newest version was v1.3.0, published 22 September 2026.

The comparison page lists Everfern as "Not scored," so you are choosing it for its features rather than for a ranking against the other agents.

Pick this if

• You want one normal app to download and open, with no Docker.
• You want an agent that can work in ordinary desktop apps the way a person does.
• You are fine with a small, young project, and you will watch it while it works.
• You want to try a local model on your own computer (Ollama or LM Studio), or bring your own key.

Skip it if you want the safest first setup. Agent Zero keeps the agent inside a container, but Everfern's computer use does not, so go back to Step 1 of the main guide instead.

Before you start

The Everfern app is free. The README says "MIT — free for personal and commercial use."

The model is a separate choice, and it is where any cost comes from, so read Pick your model and budget first. If you want the model on your own computer, check what your memory can run in Local models by hardware.

Have these ready:

• A Windows 11 PC or a Mac.
• About 400 MB of free space for the download, plus room for the app. A local model needs much more space, as the hardware page explains.
• A new, empty folder for the agent's work, which you will make in "Turn on the safety switches first" on the next page.
• About 30 minutes when nobody needs the computer.

The models Everfern supports and what they cost

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. That gives you three kinds of cost:

• A local model (Ollama or LM Studio) has no bill, but it needs a computer with enough memory, and small local models are weak at long jobs.
• A cloud key is billed by the provider that issued it, so the price depends on which provider and model you choose. A free Google Gemini key needs no card, and the full price table is in Pick your model and budget.
• EverFern Cloud is reached by signing in to the project's own service rather than with a key, since Everfern's setup screen code offers Log In to EverFern Cloud.

Everfern's website also sells paid plans called Pro and Max for "EverFern Cloud," which is the project's own hosted cloud model. You do not need them, because the Free plan includes "Unlimited local models and BYOK" (bring your own key) and a small EverFern Cloud allowance. On 14 November 2026 the pricing page listed Pro at $4.58 a month ($54.96 billed once a year) and Max at $18.33 a month ($219.96 billed once a year).

The catches

• Tiny project. It has about 101 stars and 12 forks, and the release notes this guide read all list changes from one GitHub account, so fewer people are watching for bugs.
• Most pauses are up to the model. In the v1.3.0 code, only commands on your real computer always show an approval card. The pause before destructive actions is an instruction to the model, not a lock.
• Computer use is your real desktop. The VM claim is about shell commands, while clicking and typing happen on your actual screen and apps.
• "Nothing leaves your machine" has limits. The README says that, but it is only true with a local model. A cloud model gets your requests and probably your screen pictures, and the website also sells paid EverFern Cloud plans.
• Stopping is a single button. While a job runs, the send button turns into a square stop button. It stops the job, but anything already done stays done.
• It retries by itself. The README says failed steps trigger "automatic rollback, root-cause analysis, and a retry — no babysitting required." Watch it anyway.
• Real bugs have shipped. August 2026 releases fixed the Windows-key loop and removed a voice-service key that had been written into the code.
• Big download, unclear signing. The installers are about 330 to 400 MB, and this guide could not confirm whether the Windows or Mac app is signed or notarized.
• No phone or chat-app control. The README describes a desktop app only.
• Small local models are weak. See Local models by hardware.
• Newest version is from 22 September 2026. The code was still being updated on 9 November 2026, but there was no newer download.

For other choices, see Troubleshooting and runner-ups and the landscape research.

Sources

Opened 2026-11-14.

• Everfern on GitHub and its README, for the features, models, license, and the VM and approval claims.
• Everfern latest-release API, for the tag and date.
• Everfern release list, for the v1.1.5 removed key and the v1.1.6 Windows-key loop fix.
• Everfern pricing, for Free, Pro, Max, EverFern Cloud, and the prices.
• Everfern v1.3.0 setup screen source code, for Log In to EverFern Cloud.
• Everfern v1.3.0 agent runner and tool code, for which approvals the code enforces and which are left to the model.
• Earlier research notes: other agents, Everfern section and landscape.
• research/04, for model sizes and memory.

Next

Continue to Install and first job.