⌘01 / OBSERVE
See the actions.
Follow shell commands, file changes, tool calls, network requests, and browser interactions. Keep the context across your agent workflows.
SENTINEL BY EXPLAINX · EARLY ACCESS
Let agents work.
Keep watch.
The monitoring and safety layer for everything your agents do.
For enterprise IT teams monitoring agents across the organization. For individuals who want to use agents more safely. See their actions, flag potential risks, and check AI skills when you pull them.
s sentinel / overviewCONCEPT PREVIEW
WORKSPACE / ENGINEERING
ObservingEyes on every agent.
Agent sessions03
Needs attention01
Session tokens24.8k
ACTIVITY STREAMSAMPLE SESSION
↗
Read project files
14:32:01Coding agent · filesystem.read
⌘
Run test suite
14:32:04Coding agent · shell.execute
△ Skill security checkHIGH RISK
A pulled skill contains instructions to read credentials and send them to an external URL.
skill.scan deploy-helper / SKILL.md◎
Navigate to staging app
14:32:09Browser agent · browser.navigate
Illustrative product direction. Sample activity, not a live dashboard.
Built for the way
you already work.
Claude CodeCursorCopilotCustom MCPBrowser agents
Planned integrationsEVERY ACTION DESERVES A SECOND LOOK.
Delegate the work.
Keep the context.
From a shell command to a browser click, see what your agents are doing and what needs attention. Skill security checks are part of that wider safety layer.
△02 / OBSERVE
Surface the risks.
Flag destructive commands, unexpected access, and suspicious instructions. Check AI skills for security risks when you pull them, too.
↗03 / OBSERVE
Follow the spend.
Understand token consumption by agent and session, wherever the harness exposes it. Spot the runs that need a closer look.
WHERE THIS IS GOING
Observation first.
Enforcement next.
v1 is deliberately scoped to visibility and alerts. Blocking and org-wide policy are the next builds, sequenced on purpose — not hidden.
v1 · NOWSHIPPING
Observe & alert.
Activity monitoring across shell, files, network, browser, and MCP calls. Risk flags on inconsistent or dangerous actions. Skill and MCP security scanning. Token tracking where the harness exposes it.
v2 · NEXTPLANNED
Ask & approve.
Pause on destructive commands — deleting data, reading credentials, touching SSH keys — for a human decision before they run, instead of only flagging after the fact.
v3 · LATERPLANNED
Policy enforcement.
Org-wide rules that block known-dangerous actions automatically, with SIEM / OTLP export and fleet deployment for security teams.
HOW SENTINEL COMPARES
We’re not the only
ones watching.
Agent monitoring is a real, fast-moving category. Here’s how Sentinel’s early-access scope stacks up against the closest tools, honestly — including the free ones.
| Tool | Activity monitoring | Skill / MCP scanning | Blocking | Individual tier | Price |
|---|---|---|---|---|---|
| Sentinel | Desktop + browser | Yes | Roadmapped (v2) | Planned, free | TBD |
| numbat (Perplexity) | Yes, via local hooks | Partial (MCP events) | Opt-in | Free, open source | Free |
| MintMCP Agent Monitor | Yes | Yes | Yes | Enterprise only | Custom quote |
| Legit VibeGuard | Yes | Yes | Yes | Enterprise only | Custom quote |
FOR YOUR ORGANIZATION. FOR YOU.
IT visibility.
Personal safety.
One Sentinel.
Enterprise IT teams
Monitor agent activity across your organization. Understand what ran, spot risky behavior, and follow token spend.
Individuals
Keep an eye on what your agents access, change, and send. Get potential risks flagged, including in the skills you pull.
What is Sentinel?
Sentinel is the monitoring and safety layer from explainx for agent activity: commands, files, tool calls, network requests, and browser interactions. It also checks AI skills for security risks when you pull them. Early access to the prototype is coming.
What happens when I pull an AI skill?
Sentinel runs a security check on the skill to surface potentially unsafe instructions, such as credential access or data exposure. The check helps you review what you are adding to your agent; it is not a guarantee that a skill is safe.
Which agents will it support?
We're building toward coding harnesses such as Claude Code, Cursor, and Copilot, alongside custom MCP agents and browser automation. Integration coverage is being developed; joining the waitlist helps us understand which tools to prioritize.
Will it block my agents from taking actions?
The first version is designed for observation and alerts. It does not promise to block unsafe actions or prevent every attack. Policy-based controls are planned for later versions.
Who is early access for?
Enterprise IT teams that need to monitor AI agent activity across their organization, and individuals who want a safety layer for the agents they use. Both can join the same waitlist; team details are optional.
What happens after I join?
We'll email you when early access opens. We're building the prototype now and inviting people as access becomes available. No account or payment is needed to join the waitlist.