PlatformDeployment & control

Deployment and control

CyberStrike is open source. The engine, the agents, and the tooling are all on GitHub under AGPL-3.0, so you can read the code, run it yourself, and keep every request and finding inside your own boundary. Self-host it on a laptop, in your own cloud, or fully air-gapped, or use a managed instance when you want the fastest path.

License: AGPL-3.0
Editions: Open source · Managed cloud
Runs on: macOS · Linux · Windows
Deploy: Local · Your cloud · On-prem
Models: Bring your own

Open source

The full platform is open source. Read the code, run it yourself, and keep every request and finding inside your own boundary.

Security-savvy teams don't only ask what a tool finds. They ask who reasons about the attack surface, whose infrastructure the code and findings touch, and who validates a result before it becomes a report. With CyberStrike the answer can be: you. It is published under AGPL-3.0 on GitHub, with no opaque cloud and no vendor you have to take on faith.

Install

One line to start.

The quickest way to start is the install script:

# macOS and Linux
curl -fsSL https://cyberstrike.io/install.sh | sh

# Windows (PowerShell)
irm https://cyberstrike.io/install.ps1 | iex

Or clone the repository and build from source:

git clone https://github.com/CyberStrikeus/CyberStrike
cd CyberStrike
# then follow the README to build and run

Then point it at a target and bring your own model key. CyberStrike runs from the terminal, so it fits the tooling and CI you already use, with no agents to install on the target and nothing to phone home.

Three ways to run

Self-host it anywhere, or let us run it. The same engine and the same guardrails underneath.

Option 01
Local / CLI
The open-source edition runs on your own machine. The fastest way to try it, and everything, capture, analysis, and replay, stays local.
Option 02
Your own cloud
A dedicated, single-tenant instance in an AWS, Azure, or GCP account you control, under your existing IAM and network controls.
Option 03
On-premises / air-gapped
On your hardware with no internet dependency. Paired with a local model, it needs nothing outside your network.
Option 04
Managed cloud
Hosted and fully managed, the fastest path to production and the default for a proof of value. Rolling out in early access.

Bring your own model

Model-agnostic. Your LLM, your provider agreement, or a fully offline model.

Point it at your own endpoint, your existing OpenAI, Anthropic, or Azure agreement, or an offline model (Ollama, LM Studio) for air-gapped work. What doesn't change is the harness: the guardrails, tool permissions, and policy enforcement that constrain what any agent can do sit outside the model, so swapping the model never loosens them. Substitutable models mean no single provider's outage or policy change can take you offline.

Control

Scope, budgets, permissions, and human sign-off hold in every deployment.

Explicit scope, so no tool ever sends outside the hosts you authorized; a governed send engine with budgets, rate limits, and a circuit breaker; per-agent tool permissions; and human-in-the-loop promotion, where findings land as drafts and a person signs them off. All of it is configuration that lives with your project in cyberstrike.json, version-controlled and reviewable, with a managed config that wins org-wide in an enterprise.

FAQ

Is CyberStrike really open source?

Yes. The platform is published under AGPL-3.0 on GitHub. The open-source edition is the full engine and agent system, and you can self-host it today.

Can we run it where our data can't leave our environment?

Yes. Self-host in your own cloud or fully air-gapped on-premises, pair it with a local model, and no traffic or findings leave your boundary.

Do we have to use your model provider?

No. Bring your own LLM, your existing provider agreement, or an offline model. The guardrails live outside the model, so swapping it never loosens them.