Fast answers for first-time evaluators.

Do I need to build MultiAiPanel locally?

No. The shortest evaluation path is the packaged release zip from GitHub Releases.

Is there a browser-store install today?

Not yet. 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.

What do I need before the first compare run?

A Chromium-compatible browser with Developer Mode, plus at least one supported AI chat tab already open and signed in inside the same browser profile where you load Prompt Switchboard. The empty state inside the extension now walks you through opening tabs, checking readiness, and seeding the first prompt.

Which sites are supported today?

The public support surface currently targets ChatGPT, Gemini, Perplexity, Qwen, and Grok. The supported-sites page is the canonical public list for this front door.

Does MultiAiPanel proxy prompts through a hosted backend?

No. The compare flow stays inside your browser and reuses the sessions you already trust.

Where should I go for support or onboarding?

Start with the install guide, first compare guide, supported sites page, and trust boundary page. For follow-up questions, use GitHub Discussions.

Can I export a compare result or share it outside the side panel?

Yes, but the current share surface stays export-based and host-local. Prompt Switchboard can copy a compare summary, export Markdown, and keep a readable compare artifact without turning the product into a hosted share-link service.

What are prompt packs?

Prompt packs are reusable compare starters for writing, research, coding explanations, and rewriting. They exist to get you to the first useful compare run faster than a blank input box.

Does the AI Compare Analyst change the trust boundary?

The core compare lane does not change: it still reuses the browser tabs and sessions you already trust. The AI Compare Analyst is optional and runs one structured prompt through a supported browser tab when you enable that lane.

Does this build store an analysis API key?

No. The current repository build does not ship direct browser API-key execution for the analyst lane. The shipped path uses a browser session you already trust, and the direct BYOK path stays gated until there is a safer server-backed token flow.

What does the AI Compare Analyst actually add?

It adds an optional decision-assist lane on top of the core compare board. The analyst can summarize consensus, explain disagreement, recommend the best-fit answer to continue from, and draft the next follow-up question, but it does not pretend to be an absolute automatic judge.

What does the MCP server expose?

The MCP server exposes MultiAiPanel product actions, not arbitrary browser automation. The current surface covers readiness checks, model-tab opening, compare runs, failed-model retries, session reads, compare export, the browser-session analysis lane, the optional local Switchyard runtime-backed analyst lane, and the built-in next-step workflow tools including workflow run/list/get/resume actions.

Is there a repo-local operator helper for Codex and Claude Code workflows?

Yes. Use npm run mcp:operator -- doctor, npm run mcp:operator -- server, and npm run mcp:operator -- smoke as the main maintainer path. The same helper also keeps bridge status and workflow staging inside the governed MCP tool contract instead of inventing a second public CLI protocol, while readiness returns the cached readiness snapshot and workflow-get reads the current workflow snapshot through the same local MCP server path.

Which coding agents are the main MCP bindings today?

The strongest repo-specific MCP bindings today are OpenAI Codex and Claude Code. This repo now also ships public bundle-ready MCP packets for OpenCode and OpenClaw, but those paths are still narrower and less verified than the Codex and Claude Code host lanes.

If you want the actual copy-paste setup snippets, open the MCP starter kits guide.

If you also need the public distribution truth, open the public distribution matrix.

If you want the shortest one-host packet instead of the full matrix, open the host packets page.

Where do the MCP starter files go for Codex, Claude Code, OpenCode, and OpenClaw?

Codex uses a config.toml MCP block, Claude Code uses a Claude MCP config such as .mcp.json, OpenCode uses a project-root opencode.jsonc, and OpenClaw uses openclaw mcp set or an mcp.servers entry.

The starter-kits guide and the multi-ai-panel://builder/support-matrix resource now spell out the exact placement hints.

Is MultiAiPanel already listed on an official marketplace or registry for every documented host?

No. MultiAiPanel now ships repo-owned public bundles for Codex, Claude Code, OpenCode, and OpenClaw, but that is not the same thing as an official marketplace or registry listing.

Use the public distribution matrix when you need to separate “official surface exists”, “repo-owned bundle ships now”, and “MultiAiPanel is actually published there already.”

Use the host packets page when you want the fastest exact packet for one host instead of the full matrix.

Is there a public API or SDK today?

Not yet. The current host-facing surface is the repo-owned MCP server plus the shared bridge protocol inside this repository. MultiAiPanel does not ship a public HTTP API or a general-purpose SDK in this release.

Does the MCP server need a hosted relay?

No. MCP clients connect to a local stdio server, and that server reaches the extension through a localhost bridge on the host machine.

Can I run the MCP server in Docker?

Yes. The Docker image packages the repo-owned MCP server and repo-local operator helper, but it does not turn MultiAiPanel into a hosted compare service.

The browser extension, login state, and compare-first trust boundary still stay on the host. Use the Docker integration page when you want the exact build, run, and bridge-port story.

Are workflow runs durable after the browser session ends?

No. Workflow run snapshots are session-scoped runtime cache. They are useful for the current browser session and current compare turn, but they are not a durable cold-start ledger that survives every restart. Use the workflow list/get/resume tools only while the current browser session still owns that snapshot.

Does the Switchyard runtime replace the browser-tab compare lane?

No. The optional local Switchyard runtime lane only powers the analyst execution path. MultiAiPanel still keeps the compare-first side panel, tab readiness, compare fan-out, and workflow shell inside the browser.

Does “Seed composer only” run the next compare automatically?

No. Seed-only actions only stage the next prompt in the composer. Use the workflow panel's Run next compare now action when you want MultiAiPanel to execute another compare turn through the compare-first lane.

Does this release use Chrome Native Messaging?

No. The shipped transport is a local stdio MCP server plus the localhost bridge. Native Messaging remains an optional future hardening path, not the active runtime transport in this build.

What belongs in Discussions, what belongs in Issues, and where do security reports go?

Use Discussions for onboarding, workflow ideas, and open-ended product questions. Use Issues for reproducible bugs on the supported surfaces. Use SECURITY.md for security-sensitive reports instead of posting details in public.

What if a supported site stops working after a UI change?

That is usually selector drift: the site changed its DOM and the extension needs an update. Check the latest packaged build and search Issues or Discussions before filing a new bug report.

Can MultiAiPanel tell me when a model tab is not ready?

Yes. The compare workspace now surfaces model readiness so you can see whether a selected tab is ready, still loading, missing, or likely affected by selector drift before you send the next run, and the repair center points to the next action instead of leaving you with a dead status pill.

Do I have to rerun every model when only one site fails?

No. MultiAiPanel now supports targeted recovery so you can retry the failed model cards instead of replaying the entire comparison board.

Is this a provider SDK or automation framework for any website?

No. MultiAiPanel is a compare-first browser extension for a defined set of AI chat surfaces. It is intentionally narrower than a generic automation stack, and the MCP server keeps that same product boundary instead of exposing raw browser control.