Launching OmniConnect - Authentication for Computer Use Agents

Introducing: OmniConnect

"State of the art" browser agents continue to fail teams at the enterprise. They're unreliable, require manual unblocking, face IP bans, and do not deliver on the promise of action-taking AI at scale.

That's why we're launching OmniConnect — the first end-to-end solution for agent authentication that keeps working when real-world logins get messy.

For the first time ever, computer-using agents and digital employees get an end-to-end solution for onboarding, managing, and using human-intended credentials without compromising on security and reliability.

When an AI agent needs to complete real work, it cannot fail a task because a session expired. We've spoken with teams building automation for legacy healthcare systems, complex financial services portals, and logistics freight boards. They all hit the same wall: authentication isn't a one-time gate; it's a continuous, messy reality.

Most "works fine" solutions operate on a flawed baseline. They inject a session cookie and hope for the best. On enterprise scale and standards this approach is completely non-viable — and essentially leaves browser agents outside of the hands of the people who need them the most, teams with millions of repeatable tasks a month.

The current market standard has been the biggest bottleneck in the entire browser agents stack:

• You log in once and the session continues only as long as it stays naturally signed in. If the target application forces a logout, the agent is stuck.
• Multi-step flows and mid-session multi-factor authentication prompts cause outright task failure, requiring immediate human rescue.
• Basic proxies trigger IP changes, causing platforms to revoke sessions or ban accounts instantly.
• Developers are forced to investigate and learn how to connect to each specific app, building fragile consent flows one by one.

Auth failures shouldn't break a deterministic execution loop. To scale digital employees, the underlying infrastructure must be self-healing.

Now, Anchor users finally get an end-to-end solution. OmniConnect is built to keep the agent moving even when authentication gets hard:

• Self-Healing Session Recovery: If the browser gets logged out mid-mission, Anchor can re-login automatically and seamlessly continue the task. No human rescue needed.
• Native MFA & Complex Flow Handling: We fully support authenticator-app flows and multi-step verification, designed specifically for the variability of real-world security checks.
• Integrated Anchor VPN: We eliminate the "impossible traveler" problem entirely. Consistent network routing ensures your sessions don't randomly expire due to IP anomalies.
• Zero-Configuration Credential Onboarding: Onboard customer credentials instantly for any application. We provide an auto-generated Embeddable UI, a dedicated CLI, or a live browser view — whatever fits the exact experience you want to build.

The Reality Check: “State of the Art” vs. OmniConnect

Other solutions in the market treat the browser as a dumb container for a session token. Give an agent an authenticated session, and it acts for a session. Teach an agent to authenticate itself, and it acts forever.

Unlike today's fragile agent setups, OmniConnect is built to handle the full spectrum of real-world authentication challenges without human intervention. When sessions expire and agents get unexpectedly logged out, OmniConnect auto-recovers, re-authenticates, and continues the task — no babysitting required.

MFA prompts, which typically kill a mission outright, are natively supported. Multi-step and unpredictable login flows that break conventional tools are exactly what OmniConnect was designed for.

Timeouts no longer mean failure; auth is no longer a hard stop, which means meaningfully higher task completion rates.

On the network side, Anchor VPN maintains geographic consistency to prevent the "Impossible Traveler" lockouts that basic IP-switching triggers. And unlike legacy solutions that require custom investigation for every new app, OmniConnect is zero-config — it works across any application instantly, from day one.

Get Started Today: Integrate with Agent Skills and a Dedicated CLI

We aren't just talking about the future of agent identity; we are shipping the primitives you need to build it today. Alongside OmniConnect, we are releasing:

1. A new Agent Skill to easily integrate identity handling into your AI workflows.
2. A dedicated CLI for managing identities, credentials, and metadata on the fly.
3. A Demo Website so you can test the self-healing auth experience yourself today.

This means that your onboarding will be as seamless as our browser agents. Stop letting brittle authentication break your agents.

Let's walk through how it works — from installing the CLI to extracting data from a fully authenticated browser session, without your agent ever touching a raw password or cookie.

Dive into the technical documentation here: Authenticated Applications - Anchor Docs

Install the CLI

Pick whichever package manager you prefer:

brew tap anchorbrowser/homebrew-tap && brew install anchorbrowser    # Homebrew
npm install -g @anchor-browser/cli                                   # npm

‍Then authenticate with your API key:

anchorbrowser auth login --api-key sk-your-api-key

That's it. You're ready to go.

Create an Identity

An identity is how AnchorBrowser stores the credentials it needs to log into a target application on your behalf. Think of it as a managed credential vault entry — you create it once, and every session that references it gets a live, logged-in browser automatically.
‍

anchorbrowser anchor identity create \\\\  
  --name "acme-dashboard" \\\\  
  --source "<https://app.acme.com/login>" \\\\
  --application-name "Acme Dashboard" \\\\
  --username "ops@acme.com" \\\\ 
  --password "s3cureP@ss"

You can also use --authenticator-secret for TOTP, --custom-field key=value for arbitrary fields, or --body identity.json for a full payload.

Launch an Authenticated Session

Now attach that identity to a new browser session:

anchorbrowser anchor session create \\\\
--identity-id 05eb0e98-ab13-4177-a213-4cb49ad2ae1b \\\\
  --initial-url <https://app.acme.com> \\\\
--proxy-active \\\\
  --proxy-type anchor_proxy
{
  "data": {
        "id": "93c4b8e0-8496-4b9f-95c0-8df9be97b959",
          "live_view_url": "<https://live.anchorbrowser.io?sessionId=93c4b8e0->..."
     }
}

The browser is now running in the cloud, already logged in. Open the live_view_url to watch it in real time. The session ID is cached automatically — all subsequent commands target it.

Shortcut: anchorbrowser anchor session create --interactive walks you through application lookup, identity creation, and anti-bot config in a guided wizard.

Automate the Authenticated Browser

Here's where it gets powerful. AnchorBrowser bundles Vercel's agent-browser via the proxy command — every command runs against your authenticated session over CDP, with proxy, stealth, and captcha solving handled transparently.

Script it step-by-step:

anchorbrowser proxy snapshot -i            # accessibility tree with element refs
anchorbrowser proxy click @e12             # click by ref
anchorbrowser proxy fill @e7 "Q1 revenue"  # fill a search box
anchorbrowser proxy screenshot report.png  # capture the result
anchorbrowser proxy close                  # end session

‍
Or hand it to an AI agent with a natural-language prompt:

anchorbrowser anchor session run-agent \
  --prompt "Navigate to billing, download the latest invoice as PDF, return the total" \
  --url \
  --max-steps 15

‍The agent takes over the authenticated browser, interprets the page, clicks through the UI, and returns structured output. For long-running workflows, add --async and poll with run-agent status <workflow-id>.

Or run a pre-defined task:

anchorbrowser anchor task run \
  --input "report_type=quarterly" \
  --identity-id 05eb0e98-ab13-4177-a213-4cb49ad2ae1b \
  --cleanup-sessions

‍

Teach Your Coding Agent with Skills

agent-browser ships with installable skills that teach AI coding agents (Cursor, Claude Code, Codex) specialized browser workflows. Install them and your agent activates the right skill automatically:

<pre style="background:#f5f5f5; border:1px solid #e0e0e0; border-radius:6px; padding:1.25rem 1.5rem; font-family:monospace; font-size:13px; line-height:1.8; overflow-x:auto;"><code>npx skills <span style="color:#E24B4A">add</span> vercel-labs/agent-browser --skill agent-browser
npx skills <span style="color:#E24B4A">add</span> vercel-labs/agent-browser --skill dogfood
npx skills <span style="color:#E24B4A">add</span> vercel-labs/agent-browser --skill electron
npx skills <span style="color:#E24B4A">add</span> vercel-labs/agent-browser --skill slack
npx skills <span style="color:#E24B4A">add</span> vercel-labs/agent-browser --skill vercel-sandbox</code></pre>

<pre style="background:#f5f5f5; border:1px solid #e0e0e0; border-radius:6px; padding:1.25rem 1.5rem; font-family:monospace; font-size:13px; line-height:1.8; overflow-x:auto;"><code>npx skills add vercel-labs/agent-browser --skill agent-browser
npx skills add vercel-labs/agent-browser --skill dogfood
npx skills add vercel-labs/agent-browser --skill electron
npx skills add vercel-labs/agent-browser --skill slack
npx skills add vercel-labs/agent-browser --skill vercel-sandbox</code></pre>

<table style="width:100%; border-collapse:collapse; font-size:13px; margin-top:1rem;">
  <thead>
    <tr style="border-bottom:1px solid #e0e0e0;">
      <th style="text-align:left; padding:8px 12px; font-weight:500; color:#888; width:160px;">Skill</th>
      <th style="text-align:left; padding:8px 12px; font-weight:500; color:#888;">What it teaches</th>
    </tr>
  </thead>
  <tbody>
    <tr style="border-bottom:1px solid #f0f0f0;"><td style="padding:10px 12px; font-family:monospace; font-size:12px;">agent-browser</td><td style="padding:10px 12px; color:#555;">Full API: navigate-snapshot-interact loop, auth vault, sessions, diffing, JS eval, semantic locators</td></tr>
    <tr style="border-bottom:1px solid #f0f0f0;"><td style="padding:10px 12px; font-family:monospace; font-size:12px;">dogfood</td><td style="padding:10px 12px; color:#555;">Exploratory QA — navigates like a real user, finds bugs, outputs a report with screenshots and repro videos</td></tr>
    <tr style="border-bottom:1px solid #f0f0f0;"><td style="padding:10px 12px; font-family:monospace; font-size:12px;">electron</td><td style="padding:10px 12px; color:#555;">Automate Electron apps (VS Code, Slack, Figma) via their CDP ports</td></tr>
    <tr style="border-bottom:1px solid #f0f0f0;"><td style="padding:10px 12px; font-family:monospace; font-size:12px;">slack</td><td style="padding:10px 12px; color:#555;">Browser-based Slack automation — no API tokens needed</td></tr>
    <tr><td style="padding:10px 12px; font-family:monospace; font-size:12px;">vercel-sandbox</td><td style="padding:10px 12px; color:#555;">Run inside ephemeral Vercel Sandbox microVMs with sub-second startup</td></tr>
  </tbody>
</table>

Putting It All Together

From zero to data extraction from an authenticated app — the full happy path:

# Authenticate
anchorbrowser auth login --api-key sk-your-key

# Store credentials
anchorbrowser anchor identity create \
  --name "acme-prod" \
  --source "https://app.acme.com/login" \
  --username "ops@acme.com" \
  --password "s3cureP@ss"

# Launch authenticated session
anchorbrowser anchor session create \
  --identity-id 05eb0e98-ab13-4177-a213-4cb49ad2ae1b \
  --initial-url https://app.acme.com \
  --proxy-active --proxy-type anchor_proxy

# Automate — pick your style
anchorbrowser proxy snapshot -i
anchorbrowser proxy click @e5
anchorbrowser proxy screenshot result.png

# Or let AI do it
anchorbrowser anchor session run-agent \
  --prompt "Go to Reports, download the Q1 PDF, return the total revenue"

# Clean up
anchorbrowser proxy close

The browser ran in the cloud, logged itself in, handled CAPTCHAs and bot detection transparently, and your automation never touched a raw password or cookie.

Dive into the technical documentation here: Authenticated Applications - Anchor Docs

‍