Cinematic close-up of a translucent glass CPU chip with glowing neon amber circuits. Shimmering digital particles swarm

We’ve entered the era of the 'agentic' machine. From MCP servers to AI coding assistants, our laptops are increasingly populated by autonomous agents that don't just suggest code—they execute it. But as these agents proliferate, a scary question emerges: do you actually know everything running on your system and exactly what it has permission to touch?

The Visibility Gap

For most of us, AI agents are a black box. We grant them permissions to our filesystems or API keys, and then we trust them to behave. However, the industry is waking up to a massive visibility gap. While enterprise tools like Arthur and Human Security are tackling 'agentic visibility' at the network and cloud level, the local machine has remained a Wild West.

That’s where Geiger comes in. Positioned as a "Geiger counter for AI agents," this new tool provides a read-only command that inventories every agent, plugin, and MCP server on your machine. More importantly, it translates technical permissions into plain language, telling you exactly what each agent can access.

The Danger of 'General' Access

Geiger isn't just about inventory; it's a wake-up call. The consensus among early adopters on Hacker News is clear: if Geiger finds an agent with access to your general system drive or primary work folders, you're likely using that agent wrong.

Security experts suggest that SOTA models can become unpredictable, especially as they hit their context limits. The gold standard is shifting toward strict isolation—running agents in sandboxed environments where they can't accidentally (or intentionally) wipe a directory or leak secrets via an exposed credential.

Moving Toward Agentic Governance

We are seeing a broader trend toward "Agentic Monitoring." Whether it's GitGuardian's hooks to stop secret leaks or AgentsView's cost and activity tracking, the goal is the same: moving from blind trust to verifiable observability.

As we move toward a future of fully autonomous local workflows, tools like Geiger will be essential. It's no longer enough to know that an AI is helping you; you need to know exactly where its reach ends.

Sources

Media