Give your AI agent a browser.
Local-first browser control, semantic search, and research for AI agents. No SDK. No cloud. Just a CLI.
Why OmniScout
The actuator layer for browser-using agents — same surface as Kimi WebBridge or Claude for Chrome, but you bring the model.
Install in two commands
pip install omniscout
omniscout install --skill
Then start the daemon once: omniscout daemon start
One atomic CLI per action
navigate, snapshot, click, fill, scroll, screenshot, eval, wait, tab, network, login, captcha — every verb returns structured JSON.
Sub-second calls
A long-lived daemon keeps Playwright warm. No cold-start tax on every browser action.
Stable @eN refs
Snapshot returns accessibility-tree refs that survive CSS churn — same shape Kimi WebBridge uses.
Search, answer, extract
DuckDuckGo search with local rerank, one-sentence answers, and URL-to-Markdown extraction — no browser required.
Two backends, one vocabulary
Headless Playwright by default, or drive your real Chrome via the extension — cookies and logins intact.
Fully local
No cloud API, no hosted browser, no MCP server to babysit. Your data never leaves your machine.
Agent-native output
Skill files land in Claude Code, Cursor, Codex, and Gemini skill dirs after install --skill.
~/.claude/skills/scout/
~/.cursor/skills-cursor/scout/
~/.codex/skills/scout/
Try it in 60 seconds
omniscout browser navigate https://news.ycombinator.com
omniscout browser snapshot --refs-only
omniscout browser click '@e3'
omniscout browser screenshot --out /tmp/hn.png
Drop a prompt into any agent
After
omniscout install --skill, paste one of these — your agent routes browser work through OmniScout automatically.Use OmniScout to research "open-source browser agents in 2026" and summarize
the top 5 passages with their source URLs.
Open https://news.ycombinator.com using OmniScout, snapshot interactive
elements, click the top story, and screenshot the result to /tmp/hn-top.png.
Run `omniscout browser login https://github.com --profile work`, pause for
me to authenticate, then screenshot my notifications page with that profile.
Open the Vercel pricing page with OmniScout, capture network traffic while
scrolling, then list requests matching stripe|payment|checkout.
Where OmniScout sits
These products ship the reasoning model with the actuator. OmniScout deliberately does not — bring whatever LLM you want.
| Product | Where it runs | What OmniScout borrows |
|---|---|---|
| Kimi WebBridge | Local daemon + Chrome extension | @eN refs, /status shape, optional extension backend |
| Claude for Chrome | MV3 side panel | Hybrid @eN + coordinates, click/key/scroll vocabulary |
| ChatGPT Atlas | Forked Chromium | Per-session tab groups, persistent login |
| browser-use | Playwright in-process | Backend-agnostic action surface, persistent profiles |