Independent · Open source

AI can act on your computer. See what it does.

OpenDKG builds open tools that make AI and software on your own machine easier to see, limit, and check.

Axiom launches with this site. Guardian Agent Blackbox is part of the Axiom package, built by Umanitek.

AI agent Action
Your machine Guardrail

You decide what is allowed.

A plain-language view: an agent proposes an action; a guardrail can check it; your machine and your rules remain the boundary.

The everyday question

An AI agent is software that can take actions.

It can run commands, open files, install packages, or use tools on your behalf. That makes it useful. It also means you need a way to see what it is doing and set limits.

See it

Axiom shows which applications are using the network, where they connect, and what happened over time.

Limit it

Axiom manages its own Windows Firewall rules. Optional agent guardrails add another layer of review.

Check it

Read the source, the status labels, and the stated limits. Evidence beats a promise.

Projects

Tools for visibility, boundaries, and shared context.

OpenDKG is an umbrella for several independent projects. They do different jobs. Not all of them are security tools.

OpenDKG project · Launching

Axiom

A Windows network monitor and firewall manager, with optional agent guardrails. The Axiom package includes Guardian Agent Blackbox, built by Umanitek.

Sees
Applications, endpoints, and connection history.
Can do
Manage its own Windows Firewall rules and work with optional agent guardrails.
Sends
Core monitoring stays local. Optional integrations follow their own documentation.
Does not
Replace Windows security, guarantee protection, or block every agent action.

Plain-language summary, not a certification.

See how Axiom works

Third-party · Live

Guardian Agent Blackbox

An open-source runtime guard for AI agents, built by Umanitek. It runs in audit mode by default; blocking is opt-in.

Sees
Prompts, tool calls, commands, files, packages, and skills an agent handles.
Can do
Flag risky actions and block confirmed threats when block mode is enabled.
Sends
Signed threat signals only when community sharing is enabled.
Does not
Guarantee prompt-injection immunity or stop every risky action.

Vendor-described capabilities. Blackbox is not an OpenDKG product.

How Blackbox fits

OpenDKG · Live

DKGSwarm

A shared-context connector for AI agents. Reader and writer powers stay separate, and it is not a security guard.

Explore DKGSwarm

Independent · MIT

TRACaBot

A Telegram moderation helper for spotting scam patterns. It observes by default; removal is opt-in and evidence-gated.

Explore TRACaBot

Verification and limits

What we do not claim.

The site should be useful even when the answer is “it depends.” We publish status, ownership, and limits instead of absolute promises.

No absolute protection

No tool here can promise to stop every attack or every risky agent action. Optional guardrails depend on their own rules, configuration, and the software they attach to.

What we publish

Who owns each project, what it is designed to do, what it does not do, and where the evidence lives. We do not invent customers, metrics, certifications, or partnerships.

Read the transparency page or the public governance.

Build in the open with us.

OpenDKG is an independent open-source initiative. Public project sources, governance, and contribution rules are on GitHub.