Slidev Agent

AI-powered Slidev presentation generator using Amazon Bedrock AgentCore and Strands Agents.

Features

• Automatic web research on any topic using Tavily API
• Generates Slidev-compatible Markdown presentations
• Multiple presentation styles (technical, business, educational, pitch)
• Deployable to AgentCore Runtime for serverless operation
• Slide overflow self-check: after generation, the agent calls validate_slides_fit to detect slides that are unlikely to fit within the 16:9 frame (too many lines, oversized code blocks, long lines, etc.) and automatically regenerates the offending slides until every slide is within the budget (max 3 iterations)

Requirements

• Python 3.13+
• AWS credentials configured for Bedrock access
• Tavily API key

Installation

# Clone the repository
git clone https://github.com/your-repo/slidev-agent.git
cd slidev-agent

# Install dependencies using uv
uv sync

# Copy and configure environment variables
cp .env.example .env
# Edit .env with your TAVILY_API_KEY

Usage

CLI

# Basic usage
slidev-agent "Amazon Bedrock AgentCoreの概要"

# With options
slidev-agent "Kubernetes入門" \
    --num-slides 15 \
    --style educational \
    --theme seriph \
    --output ./output/k8s.md

# All options
slidev-agent "トピック" \
    --num-slides 10 \
    --style technical \
    --theme penguin \
    --language ja \
    --output ./output/slides.md

Options

Option	Short	Default	Description
--num-slides	-n	10	Target number of slides
--style	-s	technical	Style (technical/business/educational/pitch)
--theme	-t	penguin	Slidev theme
--language	-l	ja	Output language
--output	-o	./output/slides.md	Output file path

Preview with Slidev

After generating a presentation:

# Install Slidev globally (if not installed)
npm install -g @slidev/cli

# Preview the generated presentation
slidev output/slides.md

# Or export to PDF
slidev export output/slides.md

AgentCore Deployment

Prerequisites

1. AWS CLI configured with appropriate permissions
2. AgentCore CLI installed

Deploy

# Register Tavily API key as secret
aws secretsmanager create-secret \
    --name slidev-agent/TAVILY_API_KEY \
    --secret-string "your-tavily-api-key"

# Local development
agentcore dev

# Deploy to AWS
agentcore launch

# Check status
agentcore status

Invoke (after deployment)

import boto3

client = boto3.client('bedrock-agentcore')

response = client.invoke_agent_runtime(
    agentRuntimeArn="arn:aws:bedrock-agentcore:us-east-1:123456789:agent/slidev-agent",
    runtimeSessionId="session-123",
    payload={
        "topic": "Amazon Bedrock概要",
        "num_slides": 10,
        "theme": "seriph"
    }
)

print(response['result'])

Project Structure

slidev-agent/
├── pyproject.toml          # Project configuration
├── agentcore.yaml          # AgentCore deployment config
├── .env.example            # Environment template
├── README.md
├── src/
│   └── slidev_agent/
│       ├── __init__.py
│       ├── main.py         # CLI entry point
│       ├── agent.py        # Strands Agent configuration
│       ├── runtime.py      # AgentCore Runtime handler
│       ├── tools/
│       │   ├── __init__.py
│       │   ├── search.py     # web_search, web_extract
│       │   ├── writer.py     # write_slidev_markdown
│       │   └── validator.py  # validate_slides_fit (frame overflow check)
│       └── prompts/
│           ├── __init__.py
│           └── system.py   # System prompt
├── tests/
│   └── test_tools.py
└── output/
    └── .gitkeep

Environment Variables

Variable	Required	Description
TAVILY_API_KEY	Yes	Tavily API key for web search
AWS_REGION	No	AWS region (default: us-east-1)
BEDROCK_MODEL_ID	No	Bedrock model ID (default: Claude Sonnet)

License

MIT