Appearance
Test
Status: NEEDS-ACCOUNT (GitHub clone required)
Classification: NEEDS-ACCOUNT
Lumen (the vision-first browser agent by omlabs / omlabs.xyz) is not available as a published PyPI or npm package under any obvious name. The project is GitHub-only and requires a manual clone and setup before use.
- No pip package found under
lumen,omxyz-lumen, or related names - No npm package confirmed published at
@omxyz/lumen(the package referenced in earlier SOP drafts could not be verified as installable)
The canonical source is the GitHub repository. Check github.com/omxyz/lumen or omlabs.xyz for the current install path.
What is required
Clone the repository (check the repo README for the exact URL and branch):
bashgit clone https://github.com/omxyz/lumen.git cd lumen npm installProvide a valid LLM API key (Anthropic or compatible, depending on the model configured).
Install a local browser (Chrome or Chromium).
Smoke test (after manual install from source)
Confirm the package resolves correctly after cloning and installing:
bash
node -e "require('.')" 2>&1 || echo "FAIL"If the source installs cleanly, this either prints nothing (success) or prints a module resolution success line.
Integration test: live agent run (requires API key)
Save as smoke.ts and run with npx tsx smoke.ts:
typescript
import { Agent } from "@omxyz/lumen";
const result = await Agent.run({
model: "anthropic/claude-sonnet-4-6",
browser: { type: "local" },
instruction: "Go to example.com and return the page title.",
});
console.log("Title:", result.result);Expected success output (approximate):
Title: Example DomainThis test requires:
- A valid
ANTHROPIC_API_KEYset in your environment - Chrome or Chromium installed locally
- An active internet connection
Run the built-in test suite (from source)
If you are working from the cloned GitHub repository:
bash
npm testExpected success output:
140 tests passed (~3.5s)Benchmark run (from source)
bash
npm run eval # 25 WebVoyager tasks (default)
npm run eval -- 5 # 5 tasks only (faster smoke check)Expected output: Results table showing task completion status, steps, and time per task. A clean 25/25 run takes approximately 77.8 seconds average per task.
Notes
tested=false: no published package was confirmed installable via a one-liner. The install path requires a manual GitHub clone. Commands above are documented from the repository source and were not executed in the build environment.- Verify the current published package name against the repository README before attempting install, as names may differ from earlier SOP drafts.