CouncilAI

Compare and cross-check answers from ChatGPT, Claude, Gemini, and Perplexity in a round-based multi-LLM workflow.

### ChatGPT · Claude · Gemini · Perplexity — compare and organize with one question 💡 **When hallucinations worry you, try questions like these** With a single AI, it is hard to trust dates, numbers, or citations. Examples worth running through CouncilAI: - **Tax · deductions** — "Can I deduct [expense] on my tax return as a freelancer?" - **Travel · visas · rules** — "What are the current e-visa requirements and allowed stay length for [country]?" - **Work · contracts** — "Does this employment contract clause on termination, probation, or paid leave look standard?" In R1 you see three answers **side by side**, in R2 they **cross-critique** each other, and in R2.5 Perplexity can **fact-check** further. (CouncilAI does not guarantee accuracy and is not a substitute for professional advice.) --- `CouncilAI` (Korean brand: `인공지능 원탁회의`) is a Chrome extension that ties four AI web services into one round-based flow so you can **review answers side by side in a single tab**. 🔥 **CouncilAI brings a "multi-LLM roundtable" into your browser.** Spend less time copying and pasting across tabs and **cross-check** each model's perspective inside **structured rounds (R1–R4)**. 💪 **Why use it?** 👉 **Turn multi-AI comparison into one course** ✔️ Click the extension icon → a **full-browser-tab dashboard** opens. No more fragile popups closing mid-session ✔️ R1, R2, and R3 show ChatGPT, Gemini, and Claude **side-by-side in one tab** ✔️ R2.5 fact-check opens Perplexity in a **dedicated separate window**, eliminating iframe connection refusals ✔️ **Round timeline** shows completed and in-progress stages at a glance 👉 **If it stalls, pull it forward again** ✔️ **Restart this round** — close the current round tab and reopen it fresh ✔️ **Per-bot restart** — restart only the stuck bot ✔️ **Force complete** — manually advance when a limit modal blocks response detection ✔️ **Per-bot archive tabs** — browse R1/R2/R3/R4 response history inside each bot column 👉 **After R4, you choose** ✔️ **R4 completion modal** — guides next steps based on convergence judgment data ✔️ **Optional R5** — one more cross-check round built from R4 structured summary ✔️ **Optional final synthesis** — single-bot summary document ✔️ **Start new debate** — begin a fresh question after completion 👉 **Keep your results** ✔️ Export full debate or individual rounds as **Markdown, plain text, HTML, or JSON** ✔️ Save R4 judgment data and full debate history locally 👉 **Don't miss completion while on another tab** ✔️ **Desktop notifications** when R4, R5, or final synthesis completes — click to return to the dashboard 👉 **Language is your choice** ✔️ Switch between **Korean ⇌ English** with the dropdown at the top of the dashboard 💻 **Quick start** 🔸 **Install** from the Chrome Web Store and pin it to the toolbar. 🔸 **Sign in** to **ChatGPT, Claude, Gemini, and Perplexity**. 🔸 Click the extension icon → a **full-tab dashboard** opens in a new tab. 🔸 Enter your question and click **Start ▶**. 🔸 Watch all three bots respond side by side in the R1 tab; advance when ready. 🔸 R2.5 opens a Perplexity window automatically for fact-checking. 🔸 After R4, use the completion modal to export, start R5, run final synthesis, or begin a new debate. ❓ **FAQ** 📌 **Which AI services do I need?** You should have **web accounts** for ChatGPT, Claude, Gemini, and Perplexity and be **signed in**. 📌 **Do I need a separate API key?** **No.** CouncilAI coordinates the **web service tabs** you already use. 📌 **I had a popup that kept closing — is that fixed?** Yes. v1.1 removes the popup entirely. Clicking the icon opens a **full-tab dashboard** that stays open. 📌 **R2.5 gave me a "connection refused" error. Is that fixed?** Yes. Perplexity opens as a **standalone browser window** instead of an iframe. 📌 **Claude hit its free limit and the response is stuck.** Use the **✓ Force Done** button in Claude's column header. 📌 **What are R5 and final synthesis?** **Optional follow-up steps** from the R4 completion modal. R5 runs an additional debate round; final synthesis generates a single-bot summary. **Neither is required.** 📌 **Does CouncilAI guarantee answer accuracy?** **No.** CouncilAI helps **compare, organize, and cross-check** your workflow. R4 convergence judgment is **rule-based analysis** of collected response data. Final judgment is yours. 📌 **Where does my data go?** Progress state is stored in **`chrome.storage.local`** in your browser. CouncilAI **does not send state** to a separate developer server. Export files are saved **only on your PC**. 📪 **Contact** Questions or suggestions? Email **Chartrho@gmail.com**. For the privacy policy, see **https://sites.google.com/view/councilai-privacypolicy**. --- ### Features, usage, and privacy #### B.1 Product Overview `CouncilAI` is a browser extension that ties multiple AI responses into one workflow for comparison and organization. v1.1 UI principles: - Popup removed → clicking the icon opens a **full-tab dashboard** - **Per-round independent tabs** — R1, R2, R2.5, and R3/R4 each open in a new tab - R1, R2, R3: **3-column side-by-side view** (ChatGPT, Gemini, Claude as iframes) - R2.5: Perplexity opens as a **standalone browser window** v1.1.8 additions: - **R4 completion modal** — post-R4 action guidance based on convergence judgment data - **Optional R5** — additional round from R4 structured summary - **Optional final synthesis** — single-bot summary document - **Desktop notifications** — return to dashboard when R4/R5/final synthesis completes #### B.2 Core Value 1) Stronger comparability — three model answers side by side, same tab 2) Stronger reproducibility — export prompts, responses, R4 judgment data, and full state in multiple formats 3) Operational efficiency — fewer repeat entries, precise recovery tools when automation stalls 4) Decision support — R2.5 fact-check feeds into R3 revision; R4 judgment panel enables follow-up choices (R5, final synthesis, export) #### B.3 Supported Services - ChatGPT - Claude - Gemini - Perplexity (R2.5 fact-check only) You manage each account's sign-in and access; behavior may differ by service policy, billing tier, or access limits. #### B.4 Workflow (Round Structure) | Round | Role | UI Form | |-------|------|---------| | R1 | Initial response collection | 3-column tab (ChatGPT · Gemini · Claude) | | R2 | Cross-critique and revised responses | 3-column tab (new tab) | | R2.5 | Perplexity fact-check | Control tab + Perplexity standalone window | | R3 | Fact-check acceptance, critique, revision | 3-column tab (new tab) | | R4 | Convergence judgment and response collection | Judgment panel within R3 tab | | R5 (optional) | Additional debate from R4 summary | 3-column tab (new tab) | | Final synthesis (optional) | Single-bot summary | Single-bot tab | Recovery controls are available in each round tab. #### B.5 Dashboard and User Controls **Round-level controls** - Round timeline — visualizes `R1 → R2 → R2.5 → R3 → R4` progress - Restart this round — closes the current round tab and reopens it - Next round button — activates when all bots complete **Bot-level controls** - Per-bot restart — restart only the target bot - Force complete — manually advance when response detection is blocked - Per-bot archive tabs — browse R1/R2/R3/R4 response history **After R4** - R4 completion modal — shows convergence judgment data and guides next actions - Export — Markdown / plain text / HTML / JSON - Start R5 (optional) - Start final synthesis (optional) - Start new debate The store build does **not** expose developer-only controls (force reset, debug log export, etc.). #### B.6 Exports | Format | Use case | |--------|----------| | Markdown | Notion, Obsidian, GitHub, documentation tools | | Plain text | Universal fallback, simple archiving | | HTML | Self-contained document viewable in any browser | | JSON | Structured data including round responses and R4 judgment data | - Full debate export (all rounds R1–R4) - Per-round export (current round) Files are saved to your device. CouncilAI does not upload them to any server. #### B.7 Internationalization A language dropdown in the dashboard header lets you switch between **Korean** and **English** at any time. - Applies immediately to all UI elements, export headers, and notifications. - Stored in `chrome.storage.local` and persists across sessions. Brand by locale: - Korean: `인공지능 원탁회의` - English: `CouncilAI` #### B.8 Permission Rationale | Permission | Purpose | |------------|---------| | `storage` | Settings and workflow state in `chrome.storage.local` | | `declarativeNetRequest` | Declaratively remove `X-Frame-Options` / `Content-Security-Policy` response headers for ChatGPT, Gemini, and Claude so they can be embedded as iframes in the round tab. Scope limited to `sub_frame` resource type on the four HTTPS domains. | | `notifications` | Show desktop notifications when R4, R5, or final synthesis completes while the user is on another tab, so they can click to return to the dashboard. Notifications are generated locally and not sent to an external server. | | Host permissions (HTTPS) | ChatGPT, Claude, Gemini, Perplexity origins — declarative `content_scripts` and tab coordination | | No `tabs` or `scripting` | Page injection uses `content_scripts` only | | `web_accessible_resources` | Exposes `data/selectors.json` for content scripts; `matches` limited to the same four HTTPS patterns | > **CWS review note**: You may need separate justification text for `declarativeNetRequest` and `notifications` in the permission declaration form. Use the table above. #### B.9 Privacy and Data Handling - CouncilAI runs locally in your browser. - State is stored in `chrome.storage.local`. - CouncilAI does not send state data to a separate developer server. - Export files are local downloads only. - Notification content is generated locally and not transmitted externally. Interaction with ChatGPT / Claude / Gemini / Perplexity follows each provider's policies. #### B.10 Before You Start For stable use: - sign in to all four target services beforehand, - stable network, - up-to-date browser, - keep CouncilAI tabs open while the session is running, - if your browser blocks popups, allow popups for the Perplexity domain so R2.5 can open its standalone window, - ensure OS/Chrome notification settings allow CouncilAI notifications (optional). #### B.11 Limitations and Notes - Target UI changes may require selector updates. - Speed and results may vary by rate limits, latency, or policy changes per service. - CouncilAI assists comparison and organization; it does **not** guarantee answer quality. - R4 convergence judgment is **rule-based analysis** of collected data; it does not replace your final decision. - Final decisions are yours. #### B.12 Recommended Use Cases 1) Research — collect multiple LLM views; verify facts in R2.5; export after R4 judgment 2) Draft review — cross-critique in R2, refine in R3, review in R4 3) Risk checks — spot model disagreement in R2; verify with Perplexity 4) Reporting — export the full debate history as HTML or Markdown #### B.13 Operating Summary CouncilAI does not force one model's answer as "the truth." It helps you gather and organize multiple answers so you can reach conclusions with clearer rationale. Core value: - cross-model review, - automation of repetitive work, - traceable outputs, - user-centered final judgment.