Live Assist JS-SDK

Real-time conversation intelligence — transcription, behavioral profiling, agenda tracking, and AI feedback. One Docker image with ASR, models, and agent. You only need a Gemini or Claude API key.

Quick Start (Docker)

One-liner (share with customers)

Install (auto-detects platform, installs CLI):

GEMINI_API_KEY=your_key bash -c 'curl -fsSL https://raw.githubusercontent.com/WhissleAI/live_assist_js_sdk/main/install.sh | bash'

Or run Docker directly (CPU, auto-selects amd64/arm64):

docker run -d --name live-assist -p 8001:8001 -p 8765:8765 -e GEMINI_API_KEY=your_key whissleasr/live-assist:latest

GPU (NVIDIA, faster ASR):

docker run -d --name live-assist --gpus all -p 8001:8001 -p 8765:8765 -e GEMINI_API_KEY=your_key whissleasr/live-assist:latest-gpu

Option A: Install script (recommended)

export GEMINI_API_KEY=your_key_here
curl -fsSL https://raw.githubusercontent.com/WhissleAI/live_assist_js_sdk/main/install.sh | bash

With Claude: ANTHROPIC_API_KEY=your_key LLM_PROVIDER=anthropic bash -c 'curl -fsSL ... | bash'

Bash CLI (after install):

live-assist start              # Start Docker
live-assist status             # Check health
live-assist agents             # List smart agents
live-assist feedback "We agreed to send the deck by Friday"
echo "Meeting notes..." | live-assist feedback --agent commitment_tracker
live-assist stop               # Stop Docker

Option B: Docker Compose

git clone https://github.com/WhissleAI/live_assist_js_sdk.git
cd live_assist_js_sdk

export GEMINI_API_KEY=your_key_here
docker compose -f docker/docker-compose.unified.yml up -d

Option C: Docker run

docker run -d --name live-assist \
  -p 8001:8001 -p 8765:8765 \
  -e GEMINI_API_KEY=your_key_here \
  whissleasr/live-assist:latest

GPU variant (faster ASR, requires NVIDIA GPU + nvidia-container-toolkit):

docker run -d --name live-assist --gpus all \
  -p 8001:8001 -p 8765:8765 \
  -e GEMINI_API_KEY=your_key_here \
  whissleasr/live-assist:latest-gpu

What you get

Port	Service
8001	ASR (WebSocket at `ws://localhost:8001/asr/stream`)
8765	Agent API (feedback, sessions, health)

First start takes ~2 minutes while ASR loads models. Check logs: docker logs -f live-assist.

Run the example app

Once Docker is running:

git clone https://github.com/WhissleAI/live_assist_js_sdk.git
cd live_assist_js_sdk/examples/transcript-timeline-demo
npm install
npm run dev

Open http://localhost:5173 — configure your session and click Start Live Assist.

Troubleshooting: If you see ERR_MODULE_NOT_FOUND for Vite, run rm -rf node_modules package-lock.json && npm install and try again.

Architecture

Browser ──WebSocket PCM──► ASR Server (8001)
  │                          │
  │                    transcript + metadata
  │                          │
  ◄──────────────────────────┘
  │
  ├──SSE──► Agent Server (8765)
  │           ├── Memory extraction
  │           ├── Status tracking
  │           ├── LLM feedback (Gemini/Claude)
  │           └── Action item extraction
  │
  ◄──────────────────────────┘

Packages

Package	Description
`@whissle/live-assist-core`	Framework-agnostic JS — ASR client, capture, profiling, session orchestrator
`@whissle/live-assist-react`	React components — provider, widget, donut, transcript, agenda tracker
`packages/server`	Python FastAPI agent with LangGraph workflow

Integration

React

npm install @whissle/live-assist-core @whissle/live-assist-react

import { LiveAssistProvider, LiveAssistWidget } from '@whissle/live-assist-react';
import '@whissle/live-assist-react/styles/live-assist.css';

function App() {
  return (
    <LiveAssistProvider config={{
      asrUrl: "ws://localhost:8001/asr/stream",
      agentUrl: "http://localhost:8765",
    }}>
      <LiveAssistWidget
        agenda={[
          { id: "1", title: "Discuss roadmap" },
          { id: "2", title: "Review metrics" },
        ]}
      />
    </LiveAssistProvider>
  );
}

Vanilla JS

import { createLiveAssistSession } from '@whissle/live-assist-core';

const session = createLiveAssistSession({
  asrUrl: "ws://localhost:8001/asr/stream",
  agentUrl: "http://localhost:8765",
});

session.on("transcript", (entry) => {
  console.log(`[${entry.channel}] ${entry.text}`);
});

session.on("profile", ({ user, other }) => {
  console.log("User emotion:", user.emotionProfile);
});

session.on("feedback", ({ summary, suggestions }) => {
  console.log("AI says:", summary);
});

await session.start({ includeTab: true });

// ... later
const report = await session.stop();

iframe embed

<iframe
  src="http://localhost:3001/widget?agenda=Discuss+roadmap,Review+metrics"
  width="400" height="600"
  allow="microphone; display-capture"
/>

API Reference

Core: `LiveAssistSession`

const session = createLiveAssistSession(config: LiveAssistConfig);

// Events
session.on("transcript", (entry: TranscriptEntry) => void);
session.on("profile", (profiles: { user: BehavioralProfile; other: BehavioralProfile }) => void);
session.on("feedback", (data: { summary: string; suggestions: string[] }) => void);
session.on("action", (data: { items: ActionItem[] }) => void);
session.on("memory", (data: { items: MemoryItem[] }) => void);
session.on("agenda", (items: AgendaItem[]) => void);
session.on("status", (data: { engagementScore; sentimentTrend; keywords }) => void);
session.on("error", (err: Error) => void);

// Lifecycle
await session.start({ includeTab?: boolean; agenda?: AgendaItem[] });
const report: SessionReport = await session.stop();

Core: `AsrStreamClient`

const asr = new AsrStreamClient("ws://localhost:8001/asr/stream", { metadataProb: true });
asr.onTranscript = (seg: StreamTranscriptSegment) => { ... };
await asr.connect();
asr.sendPcm(int16Array);
asr.setChannel("system");
const finals = await asr.end();

React: Components

Component	Props
`<LiveAssistProvider>`	`config: LiveAssistConfig`
`<LiveAssistWidget>`	`agenda?: AgendaItem[]`, `style?: CSSProperties`
`<TranscriptView>`	`entries: TranscriptEntry[]`, `maxHeight?: number`
`<EmotionDonut>`	`segments: { key; value }[]`, `size?: number`
`<InlineProfileChart>`	`profile: BehavioralProfile`, `size?: number`
`<AgendaTracker>`	`items: AgendaItem[]`, `compact?: boolean`
`<ProfileBadge>`	`profile: BehavioralProfile`, `label: string`
`<SessionControls>`	`isCapturing`, `onStart`, `onStop`

Server Endpoints

Method	Path	Description
POST	`/live-assist/process/stream`	SSE streaming feedback
POST	`/live-assist/session/start`	Create session
POST	`/live-assist/session/end`	End session
GET	`/live-assist/sessions`	List sessions
GET	`/health`	Health check

Configuration

`LiveAssistConfig`

{
  asrUrl: string;          // WebSocket URL for ASR server
  agentUrl: string;        // HTTP URL for agent server
  backendUrl?: string;     // Optional external backend
  deviceId?: string;       // Auto-generated if omitted
  llmApiKey?: string;      // LLM API key
  llmProvider?: "gemini" | "anthropic" | "local";
  audioWorkletUrl?: string; // Path to audio-capture-processor.js
}

Environment Variables (Server)

Variable	Default	Description
`LLM_PROVIDER`	`gemini`	LLM backend: gemini, anthropic, local
`GEMINI_API_KEY`	—	Google Gemini API key
`ANTHROPIC_API_KEY`	—	Anthropic API key
`LOCAL_LLM_URL`	—	Local LLM endpoint URL
`DB_PATH`	`./data/live_assist.db`	SQLite database path
`EMBEDDING_MODEL`	`all-MiniLM-L6-v2`	Sentence transformer model
`CORS_ORIGINS`	`*`	Allowed CORS origins
`PORT`	`8765`	Server port

CSS Theming

Override CSS variables to match your brand:

:root {
  --la-primary: #124e3f;
  --la-bg: #ffffff;
  --la-text: #1a1a1a;
  --la-border: #e5e7eb;
  --la-radius: 12px;
}

Publishing npm packages (maintainers)

The workspace ships two libraries: @whissle/live-assist-core and @whissle/live-assist-react. Publish core first, then react (react depends on the published core version).

Prerequisites: npm login with permission to publish the @whissle scope (npmjs.com or your org registry). For a private registry, set publishConfig.registry in each package’s package.json and use .npmrc for auth.

Steps:

cd live_assist_js_sdk
npm ci
npm run build

# Dry run (optional)
npm publish -w @whissle/live-assist-core --dry-run
npm publish -w @whissle/live-assist-react --dry-run

npm publish -w @whissle/live-assist-core
npm publish -w @whissle/live-assist-react

After a version bump, commit updated package.json files and tag if your release process requires it.

Development

# Build all packages
npm run build

# Build individual packages
cd packages/core && npm run build
cd packages/react && npm run build

# Run the server locally
cd packages/server
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8765

Building & Pushing (Maintainers)

Builds four images sequentially, then creates a latest manifest for auto-select:

# From live_assist repo root — builds all four
./live_assist_js_sdk/scripts/build-and-push.sh

Images:

whissleasr/live-assist:latest-amd64 — CPU (Intel Mac, Linux x86)
whissleasr/live-assist:latest-arm64 — CPU (Apple Silicon, ARM Linux)
whissleasr/live-assist:latest-gpu — GPU (NVIDIA CUDA)
whissleasr/live-assist:latest — manifest (auto-selects amd64 or arm64)

Options:

./live_assist_js_sdk/scripts/build-and-push.sh --no-push      # build only, no push
./live_assist_js_sdk/scripts/build-and-push.sh --gpu-only      # build GPU only
./live_assist_js_sdk/scripts/build-and-push.sh --amd64-only    # build amd64 + manifest (when arm64 already pushed)

Alternatively, build with Docker Compose (CPU only):

docker compose -f live_assist_js_sdk/docker/docker-compose.build.yml build

Directory Structure

live_assist_js_sdk/
├── packages/
│   ├── core/            # Headless JS library
│   ├── react/           # React UI components
│   └── server/          # Python FastAPI agent
├── docker/              # Docker images (Dockerfile.unified, Dockerfile.unified.gpu)
├── examples/            # transcript-timeline-demo, vanilla-js
├── public/              # AudioWorklet
├── package.json         # Workspace root
└── tsconfig.base.json

License

Proprietary — Whissle Inc.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Live Assist JS-SDK

Quick Start (Docker)

One-liner (share with customers)

What you get

Run the example app

Architecture

Packages

Integration

React

Vanilla JS

iframe embed

API Reference

Core: `LiveAssistSession`

Core: `AsrStreamClient`

React: Components

Server Endpoints

Configuration

`LiveAssistConfig`

Environment Variables (Server)

CSS Theming

Publishing npm packages (maintainers)

Development

Building & Pushing (Maintainers)

Directory Structure

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 17 Commits
bin		bin
docker		docker
examples		examples
node_modules		node_modules
packages		packages
public		public
scripts		scripts
.gitignore		.gitignore
BUILD_AND_TEST.md		BUILD_AND_TEST.md
CLAUDE.md		CLAUDE.md
README.md		README.md
SECURITY.md		SECURITY.md
install.sh		install.sh
package-lock.json		package-lock.json
package.json		package.json
tsconfig.base.json		tsconfig.base.json

Folders and files

Latest commit

History

Repository files navigation

Live Assist JS-SDK

Quick Start (Docker)

One-liner (share with customers)

What you get

Run the example app

Architecture

Packages

Integration

React

Vanilla JS

iframe embed

API Reference

Core: LiveAssistSession

Core: AsrStreamClient

React: Components

Server Endpoints

Configuration

LiveAssistConfig

Environment Variables (Server)

CSS Theming

Publishing npm packages (maintainers)

Development

Building & Pushing (Maintainers)

Directory Structure

License

About

Resources

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Core: `LiveAssistSession`

Core: `AsrStreamClient`

`LiveAssistConfig`

Packages