Skip to content

Acco

Acco is a voice-first web safety agent that acts as a guardian for blind and visually impaired users browsing the internet.

Code Video Cerebral Valley

Core trick

Runs a fast Haiku risk check and a deeper Sonnet analysis on each page, gating payment steps behind spoken amount readback.

Tags

accessibilitysecurityvoice

Description

Acco is a voice-first web safety agent that acts as a guardian for blind and visually impaired users browsing the internet. Users speak a goal like "Pay my electricity bill" and Acco navigates a real cloud browser, detects scams and risky pages in real-time, and guides users with expressive voice — adjusting tone, speed, and urgency based on what it sees. It features a dual-stage risk analysis pipeline (fast Haiku check + deep Sonnet analysis), domain verification to catch brand impersonation, payment confirmation gates with amount readback, and natural conversational voice responses powered by Cartesia Sonic. The agent runs on Browserbase with Stagehand for live browser automation, Claude for reasoning, and supports both push-to-talk and Cartesia Line for real-time streaming voice interaction.
adityapuranik99/acco-cartesia-browserbase-hack★ 00 forksPythonCreated Feb 7, 2026 · last push Feb 8, 2026

README excerpt

# Accessibility Co-Pilot

A voice-first web safety agent that helps visually impaired users navigate the web safely. The agent provides real-time risk assessment, natural voice guidance, and automated browser control with built-in scam detection.

## Overview

This system combines:
- **Voice Interface**: Cartesia-powered speech-to-text and text-to-speech with risk-adaptive voice profiles
- **AI Brain**: Claude-powered planning and multi-stage risk analysis (fast + deep)
- **Browser Automation**: Browserbase + Stagehand for live browser control with visual feedback
- **Safety Layer**: Multi-stage risk detection with domain verification, payment protection, and confirmation gates

## Quick Start

### Prerequisites
- Python 3.11+
- Node.js 18+
- API keys for Anthropic, Browserbase, and Cartesia

### 1. Configure Environment

```bash
cp .env.example .env
# Edit .env with your API keys
```

### 2. Run Backend

```bash
cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python main.py
```

Backend runs on `http://localhost:8000`

Optional Cartesia Line runtime (separate from current `/ws` flow):

```bash
cd backend
python line_agent.py
```

This runs the same copilot logic via Cartesia Line's real-time voice transport.
In Line mode, frontend UI state is streamed from `ws://localhost:8000/ws/ui`.

### 3. Run Frontend

```bash
cd frontend
npm install
npm run dev
```

Frontend runs on `http://localhost:5173`

## Architecture

### Backend Componen

Team