Install

curl -fsSL https://8gent.dev/install.sh | sh
8gent

This downloads the newest 8gent binary from the GitHub releases into ~/.local/bin. No npm, no root, no login. It checks the download against signed checksums before it installs, and it never replaces a newer 8gent with an older one. The script is scripts/install.sh in the 8gent-code repo, and 8gent.dev serves a pinned, reviewed copy of it byte for byte; a pull request moves the pin. If 8gent is not found afterwards, add ~/.local/bin to your PATH. To pin a version: curl -fsSL https://8gent.dev/install.sh | EIGHT_VERSION=0.20.0 sh.

On Windows, run this in PowerShell instead:

irm https://8gent.dev/install.ps1 | iex

It installs 8gent.exe to %LOCALAPPDATA%\Programs\8gent\bin and puts that folder on your user PATH. No admin rights needed.

On Debian or Ubuntu, terminal tabs need build tools installed before you install: sudo apt install python3 make g++. Without them, a terminal tab says how to enable it and exits. The bare Linux test (install, version, first screen) passed without them.

What you need

Nothing else for the binary. It is self-contained: no Bun, Node or npm needed. Bun is only for running from source.
A model, for 8gent to think with. Either a local engine such as Ollama, or an OpenRouter key with a free model (Hosted models).

Where it runs

Platform	Status
Linux (x64, arm64)	Supported through `install.sh`. Tested on a bare Debian container: it installs, `8gent --version` runs, and the first screen renders.
macOS (arm64, x64)	Supported through `install.sh`. The primary platform: the team runs 8gent on Apple Silicon Macs every day.
Windows	Supported through `install.ps1` (the PowerShell line above). 8gent.dev serves the same pinned, reviewed copy of `scripts/install.ps1` as it does for `install.sh`. To run from source instead, see Build from source.

Check it

8gent --version     # shows the installed version
8gent doctor        # engines, models, tools and config on this machine
8gent --help        # every command and flag

Get a model

With Ollama installed and running:

ollama pull qwen3:14b
8gent

On first run 8gent lists the models it found and picks the first one. Change it any time with /model.

8gent is local-first: it runs on 8gent and Ollama on your machine by default. Hosted models are added only when you add a key, for example OPENROUTER_API_KEY with a :free model. Switch providers inside 8gent with /provider. The steps are in Hosted models, and more in Models.

Build from source

The repository is public:

git clone https://github.com/8gi-foundation/8gent-code.git
cd 8gent-code
bun install
bun run tui

This runs main. It needs Bun and Git.

On Linux, some native modules compile at install time, including the one behind terminal tabs, so you need build tools and the Node 22 headers:

sudo apt install build-essential python3 mpv ffmpeg   # mpv and ffmpeg are for /dj
export npm_config_target=22.12.0 npm_config_runtime=node npm_config_disturl=https://nodejs.org/dist
bun install --frozen-lockfile
bun run tui

Update

8gent update updates 8gent from the way you installed it. Binary installs update without a confirm step, need a terminal, and never move to an older version; the binary is checked against SHA256SUMS before it is installed. A source checkout stops if you have local changes (add --force to save them to a backup branch first), only fast-forwards, shows the incoming changes, and asks y/N before it runs bun install. Inside 8gent, /update does the same and reopens your session where you left it.

If something is off

Symptom	Fix
`8gent: command not found`	Add `~/.local/bin` to your `PATH` and open a new terminal
No models found on first run	Start your engine (`ollama serve`) and pull a model, or use llama-server
The header shows the install folder, not your project	Fixed on `main` (8gent-code issue 3264). Start 8gent from your project folder
Ollama is on another machine	`OLLAMA_HOST=http://gpu-box:11434 8gent`
You want no setup questions	`8gent --yes`
You want no colour	`NO_COLOR=1 8gent`

Next: First run.

Install

On this page