Skip to content

Installation and Quickstart

AgentGuard supports Python 3.9, 3.10, 3.11, and 3.12. The product name is AgentGuard, the production PyPI distribution is agentguard-evals, and the Python import and terminal command remain agentguard.

Install from production PyPI

python -m pip install agentguard-evals
python -c "import agentguard; print(agentguard.__version__)"
agentguard --version
agentguard --help

For an isolated command installation:

pipx install agentguard-evals
agentguard --version

Do not use TestPyPI as an installation source. Its similarly named project is unrelated to AgentGuard.

Initialize an existing project

agentguard init is included in the production agentguard-evals==0.3.0 package. Preview project onboarding before writing files:

agentguard init --dry-run --ci github
agentguard init --ci github

Continue with safe project initialization for the generated-file inventory, conservative Python, Node.js, and Go detection rules, overwrite model, security boundaries, and first local and CI runs.

Run a deterministic safe evaluation

The ordinary PyPI package intentionally excludes repository examples. Clone the repository when you need benchmark fixtures, demo scripts, or committed result evidence:

git clone https://github.com/richinmrudul/agentguard.git
cd agentguard
python -m venv .venv
source .venv/bin/activate
python -m pip install -e ".[dev]"
agentguard run examples/configs/fix_auth_bug_local_command_safe.yaml --agent local-command
agentguard reports show --latest --type run

The fixture is deterministic, local, and network-free. It writes reports, command evidence, a manifest, and a trace beneath .agentguard/, which is ignored by Git.

Run the curated six-scenario showcase with:

scripts/showcase_demo.sh
.venv/bin/python scripts/showcase_metrics.py --check

Run the broader metadata-backed adversarial foundation with:

agentguard suite examples/suites/adversarial_core.yaml --allow-failures
.venv/bin/python scripts/adversarial_metrics.py --check

--allow-failures is intentional: every scenario in the current adversarial-core suite represents expected unsafe behavior.

Choose an execution boundary

  • custom-command uses the configured Docker execution path. Docker is required only for Docker-backed evaluations.
  • local-command and agent-command execute with host-user permissions and are not inherently sandboxed.
  • AgentGuard observes configured command/filesystem evidence and applies post-hoc policy checks. It is not syscall-level containment or a universal security boundary.

Continue with the demo, benchmark catalog, or real-agent evaluation guide.