Skip to content

linux.md

Latest commit

 

History

History
99 lines (74 loc) · 3.39 KB

linux.md

File metadata and controls

99 lines (74 loc) · 3.39 KB
Raw

Install FrontierAgent on Linux

Use this guide when FrontierAgent calls a hosted or already-running OpenAI-compatible endpoint. If this machine must also serve a model on an NVIDIA GPU, choose the Docker or native GPU guide instead.

Native quick start

After cloning the repository, the bootstrap helper installs uv and managed Python 3.12 when needed, configures the endpoint, and starts the native runtime:

./scripts/run-linux.sh

# Alternatives with explicit isolation requirements:
./scripts/run-linux.sh --bwrap
./scripts/run-linux.sh --docker

It does not install or run a local LLM. Use --setup-only for provisioning without starting the TUI, and put additional CLI arguments after --, for example ./scripts/run-linux.sh -- --theme mono.

For manual installation, install Python 3.12, Git, and uv, then:

git clone https://github.com/ApodexAI/FrontierAgent.git
cd FrontierAgent
uv sync --python 3.12 --extra dev
cp .env.example .env

Configure .env:

OPENAI_API_KEY=your-key
OPENAI_BASE_URL=https://your-openai-compatible-endpoint/v1
OPENAI_MODEL=your-model-name

Run:

uv run frontier-agent --mode react --cwd /path/to/project
uv run frontier-agent --mode agent_team --cwd /path/to/project

The default native runtime stores run records below <project>/.apodex/runs and caches, temporary files, dependencies, and staged inputs below <project>/.apodex/runtime/native. Commands still run with the current user's permissions, so review approvals carefully on untrusted repositories.

Native mode and your git and ssh credentials

Native mode is the default on Linux, and it redirects HOME and the XDG directories into <project>/.apodex/runtime/native so a task cannot fill your real home with caches. One consequence is worth knowing before the agent's first commit: the commands it runs do not see ~/.gitconfig, ~/.ssh, or ~/.config/gh. git commit fails with Author identity unknown, and a push over ssh finds no key.

This matches how the bubblewrap and container boundaries behave — none of them expose your home directory — so it is deliberate rather than an oversight. When you do want the agent committing on your behalf, give the repository its own identity, which lives in the workspace and is therefore visible:

git -C /path/to/project config user.name "Your Name"
git -C /path/to/project config user.email "you@example.com"

Prefer that over widening the boundary. Pushing is best left to you, outside the agent's session.

Optional isolation

Use --docker when a Docker daemon is available. For a lighter Linux-only boundary, install bubblewrap and select it explicitly:

sudo apt-get update
sudo apt-get install -y bubblewrap
uv sync --python 3.12 --extra sandbox --extra dev
uv run frontier-agent --bwrap --mode react --cwd /path/to/project

Docker access is root-equivalent on a conventional daemon. Bubblewrap depends on usable Linux user namespaces and may be disabled by a host or provider.

Windows and WSL2

Native Windows Python is not currently in the release matrix. Install WSL2, clone the repository inside its Linux filesystem, and follow this guide from the WSL shell. Docker Desktop is optional; do not assume PowerShell paths or commands are interchangeable with the Linux examples.

Return to the installation chooser.