Getting Started

Quick Start

From install to your first vulnerability report in under a minute.

Talos ships as a single Python CLI. This guide takes you from install to a real report against a local agent in under a minute. If you'd rather understand the internals first, read How Talos works.

Prerequisites

Python 3.10 or newer, with pip on your PATH.
A tool-using agent you can hit over HTTP from your machine.
An adapter that matches your agent framework — LangChain, OpenAI function-calling, Anthropic tool-use, or a small custom shim.

1. Install

Install from PyPI. We recommend a fresh virtualenv.

shell

❯ python -m venv .venv && source .venv/bin/activate
❯ pip install talos-redteam
❯ talos-scan --version
talos-redteam 0.1.0

2. Point Talos at your agent

Talos speaks to your agent through an --adapter. For a LangChain agent exposed on localhost:8000:

talos-scan

❯ talos-scan --target http://localhost:8000/agent \
            --adapter langchain \
            --severity high
→ Discovering tool graph... 4 tools found (2 high-risk)
→ Generating attack chains... 25 templates matched
→ Running Loop 1/5: indirect prompt injection via search_kb...
✓ 3 exploit classes confirmed | 2 critical, 1 high

3. Read the report

Reports are written to ./reports/ as Markdown by default. Add --pdf for a PDF export or --sarif to emit SARIF for your SIEM.

talos — output

✓ Report generated: ./reports/scan_2026-07-14.md
✓ SARIF written:    ./reports/scan_2026-07-14.sarif

Warning

Talos executes real requests against the target. Every run may issue side-effectful tool calls (emails, refunds, writes). Always point your first scan at a staging tenant.

What's next

Read the tool graph Talos discovered.
Tune severity thresholds for your risk tolerance.
Add Talos to your CI pipeline to catch regressions on every deploy.

Introduction

Installation