```
╔═══════════════════════════════════════════════════╗
║ ║
║ ✦ A I S T U D I O ✦ Bharat · Sarvam AI ║
║ ║
║ Think. Talk. Connect. ║
║ ║
║ Multi-model Council · Agent Chat · Voice AI ║
║ WhatsApp Integration · Video Gen · 40+ Personas ║
║ ║
╚═══════════════════════════════════════════════════╝
```
### 🇮🇳 Built with ❤️ in Bharat · Powered by Sarvam AI
[](#)
[](https://sarvam.ai)
[](LICENSE)
[](https://react.dev)
[](https://vitejs.dev)
[](https://firebase.google.com)
[](#-pwa--install-to-home-screen)
---
> **🔥 Free hosted AI — no API key needed.**
> Sign in with Google and instantly access **Sarvam AI** (India's own LLM) and **Hosted Ollama** — free for all users. No credit card, no setup.
---
## 📋 Changelog — Recent Updates
> All changes shipped since the last README revision.
### ✦ Video Gen — SkyReels V3 Digital Avatar Videos _(New)_
New `VideoGenPage.jsx` — full lip-sync video generation powered by SkyReels V3:
- Upload a portrait image + audio file → get a lip-synced HD 720p video in 1–3 minutes
- Audio-driven: SkyReels animates the face to match the speech in your audio clip
- Upload any local audio file (WAV/MP3) — the server converts it to a public URL automatically before sending to SkyReels (no base64 limitations)
- Free with your account — server API key used by default; bring your own SkyReels key for your own quota
- Full generation status flow: **Uploading audio → Submitting → Generating → Complete**
- Video preview panel with inline player and one-click download
- Past videos accessible via a slide-over History panel
- Cost displayed per generation (sourced from SkyReels usage data)
### ✦ AI Studio — Full App Rebrand & Multi-Page Shell
The app has evolved from a single-screen "AI Council" tool into a full **AI Studio** platform with persistent navigation and six distinct pages.
- App renamed from **AI Council** → **AI Studio · Bharat · Sarvam AI**
- New `Shell.jsx` — desktop sidebar (220px) + mobile top header + mobile bottom tab bar in one layout wrapper
- New `design.js` — shared CSS keyframes, `NAV_ITEMS` constants, `ghostBtn()` and `alertBox()` utility helpers used across all pages
- `SetupScreen.jsx` refactored into a thin router that delegates rendering to page-level components
### ✦ Six-Page Navigation
| Page | Icon | Auth Required |
| -------------- | ---- | ----------------------------------------- |
| **Home** | ⌂ | No |
| **Council** | ⚖ | Optional (managed providers need sign-in) |
| **Agent Chat** | 🤝 | Yes — shows inline sign-in wall |
| **Voice AI** | 🎙 | Yes — triggers login modal on launch |
| **WhatsApp** | 💬 | Yes — triggers login modal on launch |
| **Video Gen** | 🎬 | Yes — shows inline sign-in wall |
### ✦ HomePage — New Landing Page
Replaces the old empty default view:
- Animated headline: **Think. Talk. Connect.** with multi-color gradient shimmer
- Eyebrow badge: _Made in Bharat · Powered by Sarvam AI · 🔥 FREE_
- Floating ambient background orbs (animated radial gradients)
- 5-column feature card grid (Council, Agent Chat, Voice AI, WhatsApp, Video Gen) — each card navigates to its page
- Stats strip: **40+ Personas · 10+ Languages · 6+ Providers · Free with login**
### ✦ CouncilPage — Extracted Page Component
Council builder is now `src/components/setup/pages/CouncilPage.jsx`:
- Stage I / II / III summary cards shown at the top of the page
- Two-column layout: members on the left, templates on the right
- Category filter pills for templates (Think Tank, AI Agents, Corporate, Professional, Unfiltered)
- Inline member editing — click ✎ to expand an edit form in-place, no panel required
- Validation shows a list of incomplete members with inline **Edit →** shortcut buttons
- Templates auto-select `managed_sarvam` when the user is signed in, `ollama` when not
### ✦ AgentPage — Dedicated Agent Chat Page
`src/components/setup/pages/AgentPage.jsx`:
- Full persona grid with category filter pills at the top
- Shows an inline **sign-in wall** (not a redirect) when logged out — feature preview chips visible
- Custom personas surface in a "Mine" category automatically
- **✨ Custom Persona** action button in the page header
- Persona cards show name, tagline, badge, and a **Chat →** CTA
### ✦ VoicePage & WhatsAppPage — New Platform Pages
**VoicePage** (`pages/VoicePage.jsx`):
- Hero section with a language pill strip (10+ Indic scripts rendered natively)
- "Why it's different" feature grid: cultural context, real-time streaming, code-switching, all personas by voice
- Auth-gated launch button → opens `SarvamVoiceChat` modal after sign-in
**WhatsAppPage** (`pages/WhatsAppPage.jsx`):
- "How it works" 4-step numbered flow with color-accent step markers
- Use case grid: Morning Clarity, Deal Coach, Evening Debrief, Startup Advisor
- Auth-gated launch button → opens `WhatsAppGateway` modal after sign-in
### ✦ PageHeader — Shared Reusable Header Component
`src/components/setup/PageHeader.jsx` used by all inner pages:
- Color-matched icon box (background + border tinted to page accent color)
- `clamp()`-scaled title in Syne font
- Optional subtitle line
- Optional `extra` prop — renders action buttons right-aligned (responsive wraps on mobile)
### ✦ UserAvatar Dropdown — Direction & Positioning Fixes
- Added `dropUp` boolean prop to `UserAvatar`
- **Desktop sidebar** passes `dropUp` → dropdown opens **upward**, left-anchored, expands into main content area (not clipped by sidebar)
- **Mobile header** omits `dropUp` → dropdown opens **downward**, right-anchored, stays within viewport
- Removed the brittle CSS hack (`[style*="position: absolute"]` override) from `Shell.jsx`
### ✦ PersonaPicker — New Searchable Grouped Dropdown
Replaces the native `` for persona selection in `MemberForm`:
- Grouped by category with color-coded section headers
- Category filter pills (All + each group)
- Full-text search across persona name and system prompt text
- Chair suggestion badge (`👑 CHAIR`) shown inline on eligible personas
- 2-line prompt preview per row
- Keyboard navigable (arrow keys, Enter, Escape)
### ✦ ModelPicker — New Searchable Combobox
Replaces `` blocks from reasoning models (DeepSeek-R1, QwQ) are hidden; only the final answer is shown, with a live "🧠 Thinking…" indicator
- **Tabbed results UI** — Stage I / II / III tabs, switch freely while generation is in progress
- **Session history** — all past queries stored locally, reviewable in a full modal with per-stage tabs
- **Saved configs** — save provider + model + API key combos for quick reuse
- **Mobile friendly** — responsive layout, no white flash, safe area insets, iOS zoom prevention
### Platform Modes
| Mode | Description | Auth |
| ----------------- | -------------------------------------------------------------- | -------- |
| **⚖ Council** | 3-stage multi-model deliberation with Chairman verdict | Optional |
| **🤝 Agent Chat** | 1-on-1 conversation with 40+ expert personas | Required |
| **🎙 Voice AI** | Speak in Hindi, Tamil, and 10+ Indian languages | Required |
| **💬 WhatsApp** | AI persona living inside WhatsApp | Required |
| **🎬 Video Gen** | Lip-synced HD avatar videos from a portrait image + audio clip | Required |
---
## 🎬 Video Gen — SkyReels V3 Avatar Videos
Transform any portrait photo into a talking digital avatar. Upload a face image and an audio clip — SkyReels V3 animates the face to lip-sync with the audio and renders a 720p HD video.
### How to generate a video
1. Sign in with Google (required)
2. Navigate to **Video Gen** in the sidebar
3. Enter a prompt describing the scene (e.g. _"The person speaks affectionately to the camera. Use a static shot."_)
4. Upload a **portrait image** (or paste a public image URL)
5. Upload an **audio file** (WAV/MP3) or paste a public audio URL
6. Click **🎬 Generate Video** — the server uploads your audio, submits the job, and polls for completion
7. Your video appears in the preview panel — watch it inline or download it
### Technical details
| Spec | Value |
| ------------- | ------------------------------------- |
| Model | SkyReels V3 Single Avatar |
| Resolution | HD 720p |
| Max duration | 200 seconds |
| Architecture | DiT + Multimodal |
| Sync type | Audio-driven lip-sync |
| Audio formats | WAV, MP3 (public URL or local upload) |
| Image input | URL or base64 (local upload) |
### API key
By default your generation uses the **server's SkyReels key** — free with your account, no setup needed. To use your own quota, expand the **🔑 API Key** panel and paste your key from [api.skycoding.ai](https://api.skycoding.ai). Your key is forwarded per-request and never stored.
### Audio upload flow
Local audio files (uploaded from your device) are automatically saved to the server and served at a public URL before being submitted to SkyReels. SkyReels requires a publicly accessible HTTPS URL — it cannot process base64 data URIs directly.
```
User uploads audio
│
▼
POST /api/videogen/upload-audio
Server saves file to disk → returns public URL
│
▼
POST /api/videogen/submit
Submits {prompt, image, audio_url} to SkyReels API
│
▼
GET /api/videogen/:id/status (polls every 10s)
│
▼
GET /api/videogen/:id/result
Returns video URL + cost
```
### Backend environment variables
| Variable | Default | Description |
| ------------------- | ------------------------------------- | -------------------------------------- |
| `SKYCODING_API_KEY` | — | SkyReels API key (server default key) |
| `MEDIA_DIR` | `/media/prijak/MyDrive/html/AiStudio` | Local path where audio files are saved |
| `MEDIA_BASE_URL` | `backendurl/AiStudio` | Public base URL for saved audio files |
---
## ✦ Auth & Cloud Features
### 🔐 Google Sign-In & Firebase Authentication
Sign in with your Google account to unlock cloud features and free hosted models. Uses Firebase Auth with popup → redirect fallback — works on all browsers including iOS Safari.
- Managed providers (Sarvam AI, Hosted Ollama) require sign-in
- Session persists across tabs and page reloads
- Anonymous users continue to work with local storage and self-hosted providers
### ☁ Cloud Config Sync with End-to-End Encryption
When signed in, saved council configs sync to Firestore. API keys are encrypted with **AES-GCM 256-bit** (WebCrypto) before leaving your browser — the server only ever stores ciphertext.
- Configs sync across devices automatically
- Each saved config card shows `☁ cloud · 🔒 encrypted` badge when active
- Falls back to localStorage for anonymous sessions
### 🇮🇳 Managed / Hosted Providers (No API Key Required)
| Provider | Icon | Notes |
| ----------------- | ---- | ----------------------------------------------------------------------- |
| **Sarvam AI** | 🇮🇳 | India's own LLM (`sarvam-m`) — built for Bharat |
| **Hosted Ollama** | 🦙 | Free open-source models — mistral-nemo, deepseek-r1, llama3.1, and more |
No endpoint, no API key, no credit card. Just sign in.
### 🖥 Express Backend & Usage Analytics
Node.js/Express backend (`server.js`) with Firebase Admin auth verification, per-UID rate limiting (30 req/min default), streaming proxy, Firestore usage analytics, session logging, request ID tracing, and an admin stats endpoint.
Every inbound request is assigned a unique `REQ-{timestamp}-{counter}` ID logged alongside method, path, client IP, and token snippet — making it easy to trace a generation end-to-end in Docker logs.
---
## ✦ Previously Shipped Features
### 🌡 Temperature / Creativity Slider
| Range | Label | Effect |
| ------- | ----------- | ------------------------------------ |
| 0–35% | 🎯 Precise | Deterministic, factual, low variance |
| 35–65% | Balanced | Default — good for most questions |
| 65–100% | 🎨 Creative | More varied, exploratory, unexpected |
### 🔗 Follow-up Questions
After a verdict, ask a follow-up — the council carries the full prior verdict as context. Each round is saved in `followUpChain` and shown in History.
### ⬛ Abort / Cancel Mid-Run
Cancel at any point — partial results are preserved, a red banner is shown, and follow-up is still available.
### 📤 Export as Markdown / 🖨 PDF
`downloadMarkdown()` builds a structured `.md` report. `exportPDF()` opens a print-ready styled browser window.
### ✨ Council Templates
18 pre-built persona structures across Think Tank, AI Agents, Corporate, Professional, and Unfiltered categories — one click to load.
### 📥 / 📤 Import / Export Council JSON
Save and reload full council setups. API keys are always stripped from exports.
### 🔗 Webhook Output
Configure a webhook URL in Settings. After every completed session, AI Studio POSTs the full session JSON. Works with Zapier, Make, n8n, Slack, Notion, and Pipedream.
---
## ✦ Screenshots
Home Screen:
Configure Member:
Results View:
Manage Members:
Final Verdict:
---
## ✦ Tech Stack
| Layer | Choice |
| ------------- | ------------------------------------------------------------- |
| Framework | React 19 |
| Styling | Inline styles + CSS (no Tailwind, no CSS-in-JS lib) |
| Fonts | Syne + DM Sans (Google Fonts) |
| Auth | Firebase Authentication (Google Sign-In) |
| Database | Firestore (cloud config sync, usage analytics) |
| Encryption | WebCrypto AES-GCM 256 (client-side, before upload) |
| Local Storage | `localStorage` (sessions, configs for anonymous users) |
| Backend | Node.js + Express (managed providers, analytics) |
| Indic AI | Sarvam AI 🇮🇳 (`sarvam-m`) |
| Video AI | SkyReels V3 (`skywork-ai/skyreels-v3/standard/single-avatar`) |
| Build | Vite 6 |
| PWA | vite-plugin-pwa + custom Service Worker |
---
## ✦ Getting Started
### Prerequisites
- Node.js 18+
- One of: Ollama locally, a cloud provider API key, **or just a Google account** to use free managed providers (Sarvam AI + Hosted Ollama)
### Install & Run
```bash
git clone https://github.com/prijak/Ai-council.git
cd ai-council
npm install
npm run dev
```
Open `http://localhost:5173`.
### Firebase Setup (optional — for auth + cloud sync + managed providers)
1. Create a Firebase project at [console.firebase.google.com](https://console.firebase.google.com)
2. Enable **Authentication → Google** sign-in provider
3. Enable **Firestore** database
4. Copy your config into `.env`:
```env
VITE_FIREBASE_API_KEY=...
VITE_FIREBASE_AUTH_DOMAIN=...
VITE_FIREBASE_PROJECT_ID=...
VITE_FIREBASE_STORAGE_BUCKET=...
VITE_FIREBASE_MESSAGING_SENDER_ID=...
VITE_FIREBASE_APP_ID=...
VITE_FIREBASE_MEASUREMENT_ID=...
VITE_API_URL=https://your-backend-url.com
```
Without these, the app runs in local-only mode (no sign-in, no cloud sync, no managed providers).
### Backend Setup (optional — for Sarvam AI + Hosted Ollama + Video Gen)
```bash
PORT=8103 \
OLLAMA_BASE_URL=https://your-ollama.com \
SARVAM_API_KEY=sk-... \
SKYCODING_API_KEY=sk-... \
MANAGED_OLLAMA_MODELS=mistral-nemo:latest,deepseek-r1:latest,llama3.1:8b \
MEDIA_DIR=/path/to/your/public/html/folder \
MEDIA_BASE_URL=https://yourdomain.com/media \
ADMIN_UIDS=uid1,uid2 \
node server.js
```
Requires Firebase Admin credentials (`GOOGLE_APPLICATION_CREDENTIALS` or GCP ADC).
**Backend environment variables:**
| Variable | Default | Description |
| ----------------------- | ------------------------------------- | ------------------------------------ |
| `PORT` | `8103` | Server port |
| `OLLAMA_BASE_URL` | `backendurl` | Your Ollama instance |
| `SARVAM_API_KEY` | — | Sarvam AI API key |
| `SKYCODING_API_KEY` | — | SkyReels API key for Video Gen |
| `MANAGED_OLLAMA_MODELS` | — | Comma-separated model list |
| `MEDIA_DIR` | `/media/prijak/MyDrive/html/AiStudio` | Local path for audio file uploads |
| `MEDIA_BASE_URL` | `backendurl` | Public URL prefix for uploaded files |
| `ADMIN_UIDS` | — | Firebase UIDs with admin access |
| `ALLOWED_ORIGINS` | `*` | CORS origins |
| `RATE_LIMIT_PER_MIN` | `30` | Max requests per UID/minute |
---
## ✦ Docker
```bash
docker build -f ui_dockerfile -t ai-council .
docker run -p 8080:80 ai-council
```
Open `http://localhost:8080`. The Docker image builds the React app and serves via nginx.
**Backend Docker (with Video Gen volume mount):**
```bash
docker run -d --name=aicouncilbackend --rm -p 8103:8103 \
-e PORT=8103 \
-e SKYCODING_API_KEY=sk-... \
-e MEDIA_DIR=/media/prijak/MyDrive/html/AiStudio \
-e MEDIA_BASE_URL=backendurl \
-v /media/prijak/MyDrive/html/AiStudio:/media/prijak/MyDrive/html/AiStudio \
-v /opt/ai-council/secrets/serviceAccountKey.json:/app/serviceAccountKey.json:ro \
aicouncilbackend:latest
```
> The volume mount is required so the backend can write uploaded audio files to a directory served by your web server.
---
## ✦ Project Structure
```
ai-council/
├── public/
│ ├── logo.png
│ ├── manifest.json # PWA manifest
│ └── sw.js # Service worker
├── src/
│ ├── components/
│ │ ├── atoms/index.jsx # Spin, Badge, Toggle, TemperatureSlider
│ │ ├── setup/
│ │ │ ├── Shell.jsx # NEW — sidebar + mobile nav layout wrapper
│ │ │ ├── PageHeader.jsx # NEW — shared page header component
│ │ │ ├── design.js # NEW — shared CSS, NAV_ITEMS, helpers
│ │ │ └── pages/
│ │ │ ├── HomePage.jsx # NEW — landing page
│ │ │ ├── CouncilPage.jsx # NEW — council builder (extracted)
│ │ │ ├── AgentPage.jsx # NEW — agent chat page
│ │ │ ├── VoicePage.jsx # NEW — voice AI page
│ │ │ ├── WhatsAppPage.jsx # NEW — WhatsApp page
│ │ │ └── VideoGenPage.jsx # NEW — SkyReels V3 avatar video generator
│ │ ├── AgentScreen.jsx # Single-model agent chat (40+ personas)
│ │ ├── AuthGate.jsx # Firebase auth wrapper + Google sign-in UI
│ │ ├── DeliberationScreen.jsx # Main 3-stage pipeline orchestrator
│ │ ├── HistoryModal.jsx # Past sessions browser
│ │ ├── InstallPrompt.jsx # PWA install banner
│ │ ├── ManagePanel.jsx # Add / edit / remove council members
│ │ ├── MCPPanel.jsx # NEW — MCP server integration panel
│ │ ├── MemberCard.jsx # Member row + chairman toggle
│ │ ├── MemberForm.jsx # Add/edit member form
│ │ ├── ModelPicker.jsx # NEW — searchable model combobox
│ │ ├── PersonaCreator.jsx # Custom persona builder
│ │ ├── PersonaPicker.jsx # NEW — searchable grouped persona dropdown
│ │ ├── ResultsView.jsx # Stage I / II / III tabs
│ │ ├── SarvamVoiceChat.jsx # NEW — voice chat modal
│ │ ├── SavedConfig.jsx # Local + cloud config management
│ │ ├── SettingsModal.jsx # Webhook config
│ │ ├── SetupScreen.jsx # REFACTORED — thin 6-page router
│ │ ├── SystemPromptEditor.jsx # Inline prompt preview + editing
│ │ ├── TemplateCard.jsx # Template grid card
│ │ └── WhatsAppGateway.jsx # NEW — WhatsApp connection modal
│ ├── constants/
│ │ ├── personas.js # 40+ personas across groups
│ │ ├── providers.js # All provider configs + getAvailableProviders()
│ │ └── templates.js # 18 council templates (incl. AI Agents category)
│ ├── lib/
│ │ ├── api.js # dispatchMember() — all provider fetch logic
│ │ ├── auth.js # Firebase auth, getIdToken, apiFetch, streaming proxy
│ │ ├── cloudStorage.js # Firestore CRUD + AES-GCM encryption
│ │ ├── export.js # downloadMarkdown(), exportPDF()
│ │ ├── importExportConfig.js # parseCouncilJSON()
│ │ ├── storage.js # localStorage helpers
│ │ └── utils.js # stripThinking(), uid(), sid()
│ ├── App.jsx
│ ├── index.css
│ ├── main.jsx
│ └── styles.js
├── server.js # Express backend (managed providers + analytics)
├── firestore.rules
├── firebase.json
├── nginx.conf
├── ui_dockerfile
└── package.json
```
---
## ✦ Ollama Setup
```bash
OLLAMA_ORIGINS="*" ollama serve
```
Recommended models:
```bash
ollama pull deepseek-r1 # Reasoning — great for The Analyst
ollama pull llama3.1 # General purpose
ollama pull mistral-nemo # Efficient — good Chairman
ollama pull llama2-uncensored # Uncensored — good Contrarian
ollama pull deepseek-v2 # Good Philosopher
```
> **RTX 2070 / 8GB VRAM tip:** Stick to 7B–12B models with 4-bit quantization. Ollama queues requests sequentially — expect 5–15 min per full deliberation.
---
## ✦ Recommended Council Configurations
### 🇮🇳 Zero-Setup Indian Council (sign-in only — completely free)
| Member | Provider | Model | Persona |
| -------------- | ------------- | --------------------- | ------------- |
| The Analyst | Hosted Ollama | `deepseek-r1:latest` | Analyst |
| The Contrarian | Hosted Ollama | `mistral-nemo:latest` | Contrarian |
| The Visionary | Hosted Ollama | `llama3.1:8b` | Visionary |
| Indic Voice | Sarvam AI 🇮🇳 | `sarvam-m` | Philosopher |
| The Pragmatist | Hosted Ollama | `mistral-nemo:latest` | Pragmatist 👑 |
> No API keys. No credit card. Just sign in with Google and deliberate.
### All-Local (Ollama only)
| Member | Model | Persona |
| --------------- | ---------------------- | ------------- |
| The Analyst | `deepseek-r1:latest` | Analyst |
| The Contrarian | `llama2-uncensored:7b` | Contrarian |
| The Visionary | `llama3.1:8b` | Visionary |
| The Pragmatist | `mistral-nemo:latest` | Pragmatist 👑 |
| The Philosopher | `deepseek-v2:latest` | Philosopher |
### Hybrid (Ollama + Cloud)
| Member | Provider | Model | Persona |
| ---------------- | --------- | ------------------------- | ------------- |
| Fast Thinker | Groq | `llama-3.3-70b-versatile` | Analyst |
| Devil's Advocate | Groq | `mixtral-8x7b-32768` | Contrarian |
| Local Expert | Ollama | `deepseek-r1:latest` | Philosopher |
| The Synthesizer | Anthropic | `claude-sonnet-4-6` | Pragmatist 👑 |
### Raw Model Panel (compare models unfiltered)
| Member | Provider | Model | Persona |
| ------- | ------------ | ------------------------- | ------------ |
| Model A | Anthropic | `claude-sonnet-4-6` | Raw Model |
| Model B | Groq | `llama-3.3-70b-versatile` | Raw Model |
| Model C | OpenAI | `gpt-4o` | Raw Model |
| Model D | Sarvam AI 🇮🇳 | `sarvam-m` | Raw Model 👑 |
---
## ✦ Council Templates
Templates load a pre-built persona structure in one click. Set provider, model, and API key per member after loading.
### 🧠 Think Tank
| Template | Members | Best for |
| ------------------------ | ---------------------------------------------------- | ----------------------------------------- |
| **📊 Business Strategy** | Analyst + Contrarian + Pragmatist 👑 | Product, GTM, and operational decisions |
| **🧠 Deep Thinking** | Philosopher + Visionary + Analyst + Pragmatist 👑 | Complex, nuanced, philosophical questions |
| **⚖ Full Council** | All 5 think-tank personas | Maximum deliberation depth |
| **✨ Creative Council** | Visionary + Philosopher + Contrarian + Pragmatist 👑 | Ideation, creative strategy, design |
### 🤖 AI Agents _(New)_
| Template | Members | Best for |
| --------------------------------- | ------------------------------------------------------------------------------ | ----------------------- |
| **🔬 Research + Operator** | Deep Researcher + Web Intelligence + Red Team + Operator 👑 | Research synthesis |
| **⚗️ AI Model Debate** | Grok + Claude Ethics + Gemini + Canvas + Raw Arbiter 👑 | Model comparison |
| **⚡ Agentic Task Force** | Task Planner + Full-Stack Builder + Risk Monitor + Orchestrator 👑 | Execution pipelines |
| **🔀 Skill Pipeline** | Skill Composer + Code Architect + Systems Thinker + Executor 👑 | Modular AI workflows |
| **🦉 Socratic + Red Team** | Socratic Tutor + Red Team + Ethics Auditor + Systems Thinker 👑 | Critical reasoning |
| **🇮🇳 India Intelligence Council** | Market Intelligence + Cultural Lens + Analyst + Pragmatist 👑 | India-specific strategy |
| **🛡️ Tech Ethics Review** | Code Architect + Ethics Auditor + Risk Monitor + Red Team + Systems Thinker 👑 | Responsible tech |
### 🏢 Corporate
| Template | Members | Best for |
| ---------------------- | ---------------------------------------------------- | ------------------------- |
| **🚀 Product Launch** | CFO + CTO + CMO + Legal + CEO 👑 | Go/no-go launch decisions |
| **🌱 Startup Team** | Founder + Engineer + Designer + Growth + Investor 👑 | Early-stage strategy |
| **💼 Consulting Firm** | Strategist + Ops + Finance + Risk + Partner 👑 | Client recommendations |
### 🎯 Professional
| Template | Members | Best for |
| --------------------- | ----------------------------------------------------------- | --------------------- |
| **📰 Editorial Team** | Reporter + Editor + Legal + SEO + Editor-in-Chief 👑 | Publishing decisions |
| **🏥 Hospital Team** | GP + Specialist + Pharmacist + Ethicist + Chief 👑 | Medical case analysis |
| **⚖️ Law Firm** | Litigator + Corporate + Compliance + Associate + Partner 👑 | Legal risk assessment |
### 🔬 Unfiltered
| Template | Members | Best for |
| ---------------------- | ------------------------------------------ | ----------------------------------- |
| **🔬 Raw Model Panel** | 4× Raw Model | Model comparison, zero persona bias |
| **⚗️ Raw vs Persona** | Raw + Analyst + Contrarian + Pragmatist 👑 | Persona framing vs raw reasoning |
---
## ✦ Personas
### Think Tank (Original 5)
| Persona | Role | Best for |
| ------------------- | ----------------------------------------- | -------------------------------- |
| **The Analyst** | Structured, rigorous reasoning | Step-by-step logic |
| **The Contrarian** | Challenges assumptions, finds blind spots | Stress-testing ideas |
| **The Visionary** | Reframes from unexpected angles | Creative and strategic questions |
| **The Pragmatist** | Focuses on execution and next steps | Action plans — best as Chairman |
| **The Philosopher** | First principles, ethics, meaning | Deep or value-laden questions |
### Corporate / C-Suite
| Persona | Role |
| ---------------------- | ----------------------------------------------------------- |
| **The CFO** | Financial lens — ROI, unit economics, burn rate |
| **The CTO** | Technical feasibility, architecture risk |
| **The CMO** | Market opportunity, positioning, go-to-market |
| **The Legal Counsel** | Regulatory risk, IP, liability |
| **The CEO (Chairman)** | Final go/no-go — resolves tensions with executive authority |
### Startup Team
| Persona | Role |
| --------------------------- | --------------------------------------------------- |
| **The Founder** | Vision, product-market fit, mission integrity |
| **The Engineer** | Production reality — what works at scale |
| **The Designer** | User voice — UX, simplicity, human-centered design |
| **The Growth Lead** | Acquisition, retention, viral loops, unit economics |
| **The Investor (Chairman)** | Investment thesis — team, market, defensibility |
### Consulting Firm
| Persona | Role |
| -------------------------- | ------------------------------------------------------------- |
| **The Strategist** | Structured frameworks — Porter's, BCG, jobs-to-be-done |
| **The Operations Expert** | Execution feasibility, critical path, implementation risk |
| **The Finance Expert** | Financial modeling, unit economics, stress-testing |
| **The Risk Advisor** | Risk matrix — strategic, operational, regulatory, black swans |
| **The Partner (Chairman)** | Single client-ready recommendation across workstreams |
### Editorial Team
| Persona | Role |
| ---------------------------------- | ---------------------------------------------------- |
| **The Reporter** | Facts, evidence, what's missing, who benefits |
| **The Editor** | Clarity, narrative structure, reader impact |
| **The Legal Reviewer** | Defamation, privacy, copyright, contempt risk |
| **The SEO Strategist** | Discoverability, search intent, digital reach |
| **The Editor-in-Chief (Chairman)** | Publish, hold, or spike — the final editorial ruling |
### Hospital Team
| Persona | Role |
| -------------------------------- | --------------------------------------------------- |
| **The GP** | Holistic assessment, first-line response, red flags |
| **The Specialist** | Domain expertise, differential diagnosis |
| **The Pharmacist** | Drug interactions, dosing, monitoring |
| **The Medical Ethicist** | Autonomy, consent, beneficence, equity |
| **Chief of Medicine (Chairman)** | Final clinical management plan |
### Law Firm
| Persona | Role |
| --------------------------------- | --------------------------------------------------------- |
| **The Litigator** | Claims, defenses, litigation risk, settlement |
| **The Corporate Lawyer** | Entity structure, contracts, governance, M&A |
| **The Compliance Officer** | GDPR, FCPA, employment law, sanctions |
| **The Junior Associate** | Detail-level review — ambiguous clauses, missed deadlines |
| **The Senior Partner (Chairman)** | The firm's definitive legal position |
### AI Agent Personas _(New)_
| Persona | Role |
| ------------------------- | --------------------------------------------------- |
| **Deep Researcher** | Perplexity-style deep research synthesis |
| **Web Intelligence** | Real-time web search and current state tracking |
| **Red Team** | Exploits assumptions, finds vulnerabilities |
| **The Operator** | Converts deliberation into clean, actionable output |
| **Grok (Irreverent)** | Unfiltered, contrarian, anti-establishment takes |
| **Claude Ethics** | Safety-first reasoning, harm analysis |
| **Gemini Synthesizer** | Multi-perspective synthesis and balanced views |
| **Canvas Writer** | Long-form structured writing and editing |
| **Raw Arbiter** | No persona — pure model judgment |
| **Task Planner** | Breaks goals into executable sub-tasks |
| **Full-Stack Builder** | Implementation-focused engineering reasoning |
| **Risk Monitor** | Tracks risk, flags blockers, monitors health |
| **Workflow Orchestrator** | Coordinates multi-step AI pipelines |
| **Skill Composer** | Designs modular AI capabilities |
| **Code Architect** | System design and engineering decisions |
| **Systems Thinker** | Integration, emergent behavior, feedback loops |
| **Devin Executor** | Autonomous code execution and delivery |
| **Socratic Tutor** | Questions assumptions through Socratic dialogue |
### India-Specific Personas _(New)_
| Persona | Role |
| ----------------------- | -------------------------------------------------------------- |
| **Market Intelligence** | Indian market context, consumer behavior, regulatory landscape |
| **Cultural Lens** | Indic cultural context, regional nuance, language sensitivity |
| **Sarvam Hindi** | Hindi-native reasoning via Sarvam AI |
| **Sarvam Indic** | Indic multilingual reasoning across 10+ languages |
### Unfiltered
| Persona | Role |
| ------------- | --------------------------------------------- |
| **Raw Model** | No persona, no framing — pure model output |
| **Custom ✎** | Your own system prompt — any domain, any role |
---
## ✦ The Chairman
Receives the full transcript (all responses + reviews) and delivers one authoritative verdict. Mandate:
- Extract the **strongest** insights from each member
- **Resolve** disagreements — not average them
- Eliminate redundancy and weak reasoning
- Deliver a direct, unambiguous final answer
---
## ✦ Import / Export Council JSON
**Export:** Click **📤 Export Config**. Downloads `ai-council-config.json`. API keys are never included.
**Import:** Click **📥 Import JSON** → select a file. Council loads immediately.
```json
{
"members": [
{
"name": "The Analyst",
"provider": "managed_sarvam",
"model": "sarvam-m",
"personaLabel": "The Analyst",
"systemPrompt": "...",
"isChairman": false
}
]
}
```
---
## ✦ Webhook Output
Configure a URL in **⚙ Settings**. After every session, AI Studio POSTs:
```json
{
"type": "session_complete",
"ts": 1234567890,
"query": "your question",
"temperature": 0.7,
"memberNames": ["The Analyst", "The Contrarian", "..."],
"responses": { "memberId": "response text" },
"reviews": { "memberId": "review text" },
"verdict": "Council's Verdict: ..."
}
```
Works with Zapier, Make, n8n, Slack, Notion, Pipedream. Use **🧪 Send Test Ping** to verify first.
---
## ✦ Think Block Stripping
Reasoning models (`deepseek-r1`, `qwq`) emit `... ` blocks. AI Studio strips these automatically and shows a `🧠 Thinking deeply…` indicator mid-stream. Peer review and Chairman prompts only receive the cleaned output — no thinking noise leaks between stages.
---
## ✦ Keyboard Shortcuts
| Shortcut | Action |
| ----------- | ------------------------ |
| `⌘ + Enter` | Submit query / follow-up |
---
## ✦ Privacy & Data
- **Managed providers (Sarvam AI, Hosted Ollama):** Requests are proxied through the AI Studio backend. Provider, model, stage, and character counts are logged for usage analytics. Raw prompt text is not stored server-side.
- **Self-hosted providers:** All API calls go directly from your browser to the provider. Nothing passes through our backend.
- **Cloud configs:** API keys are AES-GCM encrypted in your browser before upload. The server only stores ciphertext.
- **Local storage:** Session history and configs for anonymous users stay in your browser only.
- **API keys** are never included in JSON exports.
- **Webhook** is opt-in — nothing is sent without you setting a URL.
- **Video Gen audio uploads:** Local audio files are saved to the server's media directory and served publicly so SkyReels can fetch them. Files are named `audio_{uid}_{timestamp}.ext` — not guessable, but not private. Use a public audio URL directly if you need privacy.
---
## ✦ Limitations
- Ollama models run **sequentially** (one at a time per endpoint) — cloud models run in parallel
- No markdown rendering in responses — plain text only (`pre-wrap`)
- Session history is capped at the last 30 sessions
- PDF export uses browser print — use Markdown export for richer formatting
- Managed providers require a signed-in Google account
- Video Gen requires a public HTTPS URL for both image and audio — local file uploads are handled automatically by the backend
---
## ✦ Roadmap
- [ ] Markdown rendering in response panels
- [ ] Web search injection for grounding answers in current data
- [ ] Token usage / cost tracking per session
- [ ] Named council presets — save full council with provider + model assignments
- [x] ~~WhatsApp Integration~~ — shipped ✓
- [x] ~~Voice Mode~~ — shipped ✓
- [x] ~~Video Gen~~ — shipped ✓
- [ ] Document Council — upload a PDF and let the council analyse it
- [ ] Public Persona Gallery — publish and browse community-created personas
- [ ] MCP tool use in Council mode — let members call external tools mid-deliberation
- [ ] Video Gen history gallery — browse and replay past generated videos
---
## ✦ License
MIT — do whatever you want with it.
---
🇮🇳 **Built with pride in Bharat · Powered by [Sarvam AI](https://sarvam.ai)**
_The council is always in session._