Skip to content

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

7 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Lens Agents lab

An interactive 3D guide to how Lens Agents runs AI agents in governed sandboxes on Kubernetes. It follows one OpenClaw sandbox from nexusctl sandbox create to the audit trail.

It is plain HTML and ES modules with three.js from a CDN. There is no build step.

Run it locally

ES modules do not load from file://, so serve the folder over HTTP:

python3 -m http.server 8000

Open http://localhost:8000. The guided tour starts by itself. To open one chapter directly, add its id to the URL, for example http://localhost:8000/#ingress.

Chapters

# id What it shows
1 overview The control plane outside the cluster, a Kubernetes cluster of Kata microVM Pods, the internet, and a private network behind a firewall
2 boot Reconciler → Pod → scheduler → Kata microVM → init containers → supervisor → policy frame → OpenClaw as uid 1000
3 ingress Browser → wildcard ingress → host match → Lens ID sign-in and session cookie → data tunnel → 127.0.0.1:18789
4 network nftables cage, host-name policy, direct and upstream transport, NXDOMAIN, proxy bypass attempts
5 policy The org ceiling and project grants, drift, pg_notify propagation, the emergency halt
6 credentials Write-only credentials, __lens_cred:<id>__ placeholders, TLS interception and header injection
7 inference The LLM proxy: halt, policy, budget, PII masking, provider, metering
8 mcp The MCP gateway: attach upstreams (OAuth or through a cluster relay), the tool catalog, allowedTools
9 rbac Principals, roles, and the audit trail by source

Keys

1–9 open a chapter. ← / → go to the previous or next chapter. T starts or stops the tour. V turns the narrator on or off: the browser reads each tour caption aloud. Space pauses. L shows or hides labels. M toggles the tilt-shift effect. / hides the UI. R resets the camera. ? opens help.

Files

  • index.html: layout, styles, and the help dialog
  • world.js: the scene (cluster, sandbox Pod, control plane, internet, private network, pipes)
  • openclaw.js: the OpenClaw lobster mascot
  • app.js: the simulation (packets, policy evaluation, chapters, panels, tour, audit trail)
  • og.jpg: the 1200×630 link preview image

Accuracy

The lab follows the nexus-monorepo sources. It simplifies layout and timing, and it does not simplify behaviour:

  • The in-sandbox parts come from lens-sandbox-core, the same core that lns uses: the supervisor, the nftables cage, the proxy, the DNS stub, and placeholder injection.
  • Nexus has no approval (ask) flow, so the lab shows only allow and deny. A host with no rule gets networkDefaultVerdict.
  • Policies are org-scoped or project-scoped. There is no team policy tier.
  • The control plane is drawn outside the cluster to make the picture clearer. The Helm chart can also run it inside the cluster.

Testing hook

window.__lab exposes go(i), advance(seconds), request(key), openUI(), askModel(), the state S, and current. advance steps the simulation without requestAnimationFrame, so headless browsers can drive it.

Credit

Built on the design and code of lns-lab.

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages