First compare guide

Reach one real compare run, then repair the exact blocker instead of guessing.

This page is the shortest public map from “the extension is installed” to “the first compare turn really worked.” Think of it like a boarding checklist before takeoff: one browser, one profile, supported tabs open, then a clear repair ladder if a selected model is still blocked.

The supported install path today is the packaged GitHub Release zip. Browser-store submission materials are being kept ready, but GitHub Releases remains the supported install surface today.

Install

Load the packaged build in the same browser profile where your trusted AI tabs already live.

Compare once

Refresh readiness, send one real prompt, and wait for one board instead of five disconnected tabs.

Keep the turn

Repair only the blocked model, then export or continue from the same compare turn.

What success looks like

Before you press send

The extension is loaded in a Chromium-compatible browser.
The supported AI sites you want to compare are already open.
Those tabs are signed in inside the same browser profile.
The side panel can see the selected tabs as ready.

After the first good run

You see one compare board instead of bouncing between tabs.
You can copy a compare summary or export Markdown.
You can retry only failed models instead of replaying everything.
You can keep the workflow in the browser without a hosted relay.

First compare checklist

Install the packaged release from the latest GitHub Release.

Open chrome://extensions, enable Developer Mode, and use Load unpacked on the extracted folder.

If the MultiAiPanel icon is hidden, open the browser Extensions menu, pin it, and click the toolbar icon to open the side panel.

Open the AI tabs you want to compare in the same browser profile.

Use the built-in readiness action before the first compare run.

Send one compare prompt and watch the answers line up in the same board.

Repair ladder when the first turn is blocked

Tab not open yet

Open the missing supported AI site in the same browser profile, then run readiness again.

Still signing in

If the site still shows sign-in or sign-up UI, treat that as a login-state blocker first. Fix the human login in that same tab, then rerun the compare turn.

Readiness says blocked

Use the side panel readiness details to see whether the tab is missing, loading, mismatched, or not exposing send controls yet. That tells you what to repair next.

Coding-host confusion

The MCP server is optional. You do not need Codex, Claude Code, OpenCode, or OpenClaw just to get the first compare run working.

Real Chrome proof lane

Official Google Chrome branded builds 137+ / 139+ no longer reliably auto-load unpacked extensions from command-line flags. Keep automated runtime proof on Chromium or Chrome for Testing. If you are proving the flow in a real Chrome profile, open chrome://extensions, enable Developer Mode, and use Load unpacked manually in that same signed-in profile.

A login-gated tab is not the same thing as a product bug. In plain English: if the site itself is still asking the human to sign in, MultiAiPanel cannot honestly treat that tab as compare-ready yet.

After the first compare works

Stay in the product lane

Copy the compare summary, export Markdown, retry only the failed models, or seed the next compare turn from the strongest answer without leaving the side panel story.

Optional coding-host route

If you already use a coding agent that speaks MCP, move next to the MCP agents page or the starter kits only after the first compare works. That route is secondary, not required for the first compare.