One prompt, many AI chats, one side panel.
Teach an agent or operator one browser-first compare workflow: check readiness in already-open AI tabs, ask once from the side panel, then retry or export from the same turn.
Prompt Switchboard is a compare-first, browser-native AI compare workspace. It lets you send one prompt to ChatGPT, Gemini, Perplexity, Qwen, and Grok, then compare the replies in one side panel instead of bouncing between tabs.
After the core compare flow is clear, the repo also exposes governed integrations for Codex and Claude Code browser workflows. OpenCode and OpenClaw remain later, packet-style follow-through lanes instead of the main product story.
Agent-facing truth comes after the product story is clear: the first thing this repo teaches is how to run one real compare turn inside the extension itself. Registry packs, host packets, and public bundles are supporting surfaces around that browser product, not the first install or first success door.
Trust boundary
Prompt Switchboard stays inside your browser, uses your existing sessions on supported sites, and does not add a hosted relay or account layer. The supported repo build also does not rely on OS-level desktop automation, Force Quit helpers, or host-wide process cleanup.
Install the latest build • Landing page • Install guide • First compare guide • Supported sites • Trust boundary • FAQ guide • Privacy • Security • Building locally
The shortest way to evaluate Prompt Switchboard is simple: install the latest packaged build, keep the AI tabs you already use open, then ask once from the side panel and compare the answers in one place.
| Surface | Current truth |
|---|---|
| Product install | GitHub Release zip is the supported install path today. |
| Core product | Browser extension + compare-first side panel with your existing signed-in tabs. |
| Optional integrations | Governed coding-agent integrations for Codex and Claude Code, plus repo-owned starter packets for OpenCode and OpenClaw, and an optional Docker wrapper for the same surface. |
| Official registry | The official MCP Registry already returns a live Prompt Switchboard MCP entry for the governed integration surface. |
| Not live yet | Browser store, host marketplaces, and any Glama listing. |
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. The official MCP Registry already returns a live Prompt Switchboard MCP entry for the same governed integration surface. That registry proof does not make the browser-store path live, and it does not mean every host marketplace is already published.
The optional integration lane stays one step lower in the information hierarchy. Reach for MCP starter kits, host packets, Docker integration docs, and distribution truth only after the first compare path is already clear.
Before the first compare run, make sure the supported AI tabs you want to use are already open and signed in inside the same browser profile. The side panel now includes a first-run checklist and readiness repair actions, so the shortest path to success lives inside the product instead of only in the docs.
If you only remember one route through this repo, remember this one:
- Install the latest build from GitHub Releases.
- Run one real compare from the side panel with the tabs you already trust.
- Stay in the same turn to retry failures only or export a readable compare artifact.
Use these pages in that exact order:
- Compare responses side by side: keep the same prompt aligned across multiple model cards instead of bouncing between tabs.
- Check readiness before you send: see which selected model tabs are ready, still loading, missing, or likely affected by selector drift.
- Repair blocked models without guesswork: readiness now points you to the next action when a tab is missing, loading, mismatched, or not exposing the send controls.
- Recover only the failures you care about: retry the models that failed instead of replaying the whole compare run.
- Turn disagreement into the next move: seed the next compare round, keep seed-only actions honest, and run the next compare only when you choose to.
- Carry useful results outside the side panel: copy a compare summary, export Markdown, or keep a readable local artifact instead of only a backup dump.
- Add optional AI analysis without replacing the core compare lane: the AI Compare Analyst can summarize consensus, explain disagreement, recommend the best-fit answer to continue from, and draft the next question by reusing one browser tab you already trust, while the main compare flow stays inside the same browser workspace.
- Expose product actions through governed coding-agent integrations: Prompt Switchboard can expose readiness, compare, retry, export, session, analyst, and next-step workflow actions to local agents without becoming generic browser automation.
- Keep everything local in your browser: no hosted relay sits between your prompt and the supported AI sites.
- Reuse the AI tabs you already use: Prompt Switchboard works with the browser sessions you already keep open.
- Start from reusable prompt packs: launch writing, research, coding, and rewriting compare runs without starting from a blank prompt every time.
- Export, restore, and reuse compare runs: carry compare runs between machines through local import/export and save repeatable prompt recipes.
Before you start:
- a Chromium-compatible browser with Developer Mode available
- at least one supported AI chat tab already open and signed in
- Open the latest Releases page.
- Download the packaged extension zip, unzip it locally, open
chrome://extensions, enable Developer Mode, and use Load unpacked on the extracted folder. - If the Prompt Switchboard icon is hidden, open the browser Extensions menu, pin Prompt Switchboard, then click the toolbar icon to open the side panel.
- Open the supported AI tabs you want to compare, then ask once from the side panel.
Today the public install path is the packaged GitHub Release zip. A lower-friction store distribution path is being prepared, but it is not live yet.
If you are validating the real Chrome proof lane, keep one extra rule in mind:
official Google Chrome branded builds 137+ / 139+ no longer reliably auto-load
unpacked extensions from command-line flags. Automated runtime proof should use
Chromium or Chrome for Testing. Real Chrome proof keeps the same signed-in
profile, then uses chrome://extensions -> Developer Mode -> Load unpacked
manually.
Need the local build path, release workflow, Docker integration lane, or front-door
maintenance steps? Read CONTRIBUTING.md and the dedicated
Docker integration page.
Maintainer-only cleanup and runtime hygiene commands stay in
CONTRIBUTING.md so this README can stay focused on the
public product surface.
If you want to see the value quickly, try one of these on three or more supported sites:
Summarize the launch plan for a browser-native AI extension in three bullets.Compare the trade-offs between React and Vue for a browser extension UI.Rewrite this paragraph in a clearer, friendlier tone for a GitHub README.
- Compare ChatGPT vs Gemini vs Perplexity
- Best AI for rewriting text
- Best AI for coding explanations
- Why browser-native AI comparison matters
- Prompt packs
- Retry only the failed cards from the same compare turn.
- Export a readable compare summary or Markdown artifact.
- Reuse Prompt packs when you want a faster second run.
If you already use MCP-capable coding agents, come here after the first compare works:
- Prompt Switchboard for Codex, Claude Code, and MCP agents
- Prompt Switchboard MCP starter kits
- Prompt Switchboard Docker integration
- Prompt Switchboard host packets
- Prompt Switchboard public distribution matrix
These builder surfaces are intentionally second-ring pages. They are real and useful, but they should not outrank install, first compare, supported sites, and trust boundary in the first impression.
If you need standalone skill folders for host-specific submission flows, use public-skills/README.md. Those packets are repo-owned submission materials for OpenHands/extensions and ClawHub-style publish flows; they are not proof that any public listing is already live.
The strongest product claim here is not abstract AI productivity. It is much simpler: Prompt Switchboard removes the messy part of side-by-side comparison.
| Manual multi-tab compare | Prompt Switchboard |
|---|---|
| Paste the same prompt into every site | Ask once from the side panel |
| Wait in separate tabs and windows | Watch status chips update in one board |
| Reconstruct which answer belongs to which model | Keep aligned model cards in one compare view |
| Copy results back into your own notes by hand | Copy the best-fit answer or reopen the original tab directly |
| Lose the comparison context after the session | Keep the run saved locally for export and restore |
- Open the sites you already use: keep ChatGPT, Gemini, Perplexity, Qwen, or Grok signed in inside normal browser tabs.
- Ask once from the side panel: Prompt Switchboard fans the same prompt out from one local workspace.
- Compare clearly: review the answers side by side, inspect the per-model run timeline, copy the best response, or jump back into the original model tab.
- Recover, export, and continue: retry only the failed models, use the repair center when readiness blocks a run, export a readable compare artifact, or seed the next compare round from the completed answers.
Prompt Switchboard also includes governed coding-agent integrations for product-level workflows. Those integrations are real, but they are not the default first-stop story of the repo. The default story is still: install, run one compare, then export or retry from the same turn.
- The integration speaks MCP over
stdio. - The extension bridge stays on your own machine at
127.0.0.1. - The exposed surface stays product-specific: readiness, compare, retry, export, session reads, the analyst lane, and workflow helpers.
- The optional Docker integration wraps the same local MCP surface; it does not turn Prompt Switchboard into a hosted compare service or public HTTP API.
- The MCP surface does not expose arbitrary DOM selectors or generic website automation.
Current truthful split:
- Supported now: Codex and Claude Code are the strongest repo-specific host flows.
- Official MCP Registry: a live Prompt Switchboard entry already points to the governed integration install/docs surface.
- ClawHub: the standalone compare-workflows skill is now publicly listed at
https://clawhub.ai/xiaojiou176/prompt-switchboard-compare-workflows. - awesome-opencode: the curated project/resource entry is submitted on
awesome-opencode/awesome-opencode#276and is still review-pending; that receipt does not make the OpenCode plugin package published. - Packet-ready, not published: OpenCode still stays on the repo-owned packet lane, and this repo still does not claim a broader official OpenClaw plugin or marketplace listing beyond the live ClawHub skill page.
- Still external-only: package namespace control, browser-store submission, and any future marketplace or registry listing.
Use the repo-local operator helper for the main maintainer path:
npm run mcp:operator -- doctor
npm run mcp:operator -- server
npm run mcp:operator -- smoke
npm run mcp:operator -- live-probe
npm run mcp:operator -- live-diagnose
npm run mcp:operator -- live-support-bundleUse these links instead of keeping the full builder ledger duplicated in the README:
- MCP agents guide
- Host packets
- Public distribution matrix
mcp/integration-kits/support-matrix.jsonmcp/integration-kits/public-distribution-matrix.json
The machine-readable builder truth lives at
prompt-switchboard://builder/support-matrix.
Quick placement map:
- Codex ->
config.toml - Claude Code ->
.mcp.json - OpenCode -> project-root
opencode.jsonc - OpenClaw ->
openclaw mcp setormcp.servers
If host wiring looks correct but site behavior still feels brittle, read
prompt-switchboard://sites/capabilities next. That resource is the current
per-site DOM/readiness/private-API boundary map for the compare-first product
surface.
Native Messaging is not the shipped transport in this release. If you want
to explore that direction later, start from the scaffold notes in
mcp/native-messaging/README.md instead of
treating it as an already-wired runtime path.
The demo now shows the actual product rhythm: ready state, compare fan-out, workflow staging, and a completed comparison board.
This detail view highlights the compare-first design with the current next-step lane: one prompt header, WorkflowPanel, analyst guidance, clear model identity, delivery status chips, and direct links back to the original site.
The workflow map makes the runtime boundary explicit: Prompt Switchboard orchestrates the browser-side flow, while the supported AI websites remain the actual execution surfaces.
Settings keep the project honest as a real tool, not just a hero screenshot: export and import, language, theme, and keyboard preferences all live inside the extension.
- ChatGPT
- Gemini
- Perplexity
- Qwen
- Grok / xAI
These integrations depend on live DOM structure. When a supported site changes markup, Prompt Switchboard may need selector updates before the compare flow fully recovers.
Need the public-facing install and support detail page? Read docs/supported-sites.html.
Good fit
- You already use multiple AI chat sites and want a faster way to compare answers.
- You want the trust boundary to stay inside the browser instead of adding another hosted layer.
- You want session history and settings to stay in the same browser workspace.
Not the goal
- A cloud dashboard that proxies prompts through a backend.
- A provider-neutral SDK for arbitrary model APIs.
- A browser automation framework for non-supported websites.
- A generic AI chat app that replaces the compare-first browser workflow.
Use the public support pages for the shortest answers:
The short version is still:
- no hosted relay
- no browser-store install today
- no public SDK or generic browser automation claim
Use the public issue tracker for non-sensitive bugs, setup questions, or product feedback:
https://github.com/xiaojiou176-open/multi-ai-sidepanel/issues
For security-sensitive reports, follow SECURITY.md instead of opening a detailed public issue.
For open-ended product ideas, workflow discussion, or compare-first feedback, use GitHub Discussions:
https://github.com/xiaojiou176-open/multi-ai-sidepanel/discussions
Track packaged builds and release notes on the Releases page.
If Prompt Switchboard makes multi-model comparison easier for you, star the repo so the latest packaged builds, selector drift fixes, and compare-first front-door updates stay easy to find.