ExaFlow: Advanced Semantic Search & AI Integration

What is ExaFlow?

Semantic search toolkit for AI applications. Built with TypeScript, ExaFlow combines Exa's neural search with intelligent caching, real-time streaming, and MCP integration.

Perfect for:
- 🤖 AI applications with enhanced search
- 🔬 Research automation & multi-step pipelines
- 📰 Content analysis with live crawling
- 🛠️ Developer tools & CLI workflows
- 🏢 Enterprise search with advanced filtering

Installation

$3

``bash
bun install -g exaflow
`

$3

`bash
npm install -g exaflow
`

$3

`bash
exaflow --version
`

Setup

Create a .env file with your Exa API key:
`env
EXA_API_KEY=your_exa_api_key_here
`

Usage

$3


Get code & technical content with token limits.
`bash
exaflow context "React hooks patterns" --tokens 3000
exaflow context "TypeScript patterns" --tokens 5000
`

$3

Semantic search with neural ranking.
`bash
exaflow search "machine learning trends 2024" --type neural --num-results 20
exaflow search "AI research papers" --type neural --include-contents
`

$3

Extract content from URLs with live crawling.
`bash
exaflow contents --ids urls.txt --livecrawl always --subpages 5
exaflow contents --stdin --subpage-target "about,news"
`

$3

Async search with enrichment & polling.
`bash
WEBSET_ID=$(exaflow websets create --output json | jq -r '.data.webset.id')
exaflow websets search --webset-id $WEBSET_ID --search-query "AI research"
exaflow websets poll --webset-id $WEBSET_ID
`

$3

Multi-step research with structured output.
`bash
exaflow research --instructions "Analyze latest AI trends" --poll
exaflow research --instructions-file prompt.md --schema output-schema.json
`

$3

Integrate with AI applications.
`bash
exaflow mcp-server
exaflow mcp-server --transport http --port 3000
exaflow-mcp # Global binary
`

Options

`bash
-c, --concurrency # Parallel operations (1-20, default: 5)
-t, --timeout # Request timeout in ms (default: 30000)
--compact # Compact JSON output
--silent # Suppress event streaming
--output # Output format: json|text (default: json)
`

Architecture

`
src/
├── clients/ # Exa API clients
│ ├── base-client.ts
│ ├── exa-context.ts
│ ├── exa-search.ts
│ ├── exa-contents.ts
│ ├── exa-websets.ts
│ └── exa-research.ts
├── util/ # Utilities
│ ├── concurrency.ts
│ ├── http.ts
│ ├── http-cache.ts
│ ├── streaming.ts
│ └── fs.ts
├── cli.ts
├── mcp-server.ts
└── schema.ts
`

Response Format

All commands return structured JSON with status, timing, citations, and data:

`json
{
"status": "success|partial|error",
"taskId": "unique-task-id",
"timing": {
"startedAt": "2024-01-01T00:00:00Z",
"completedAt": "2024-01-01T00:01:00Z",
"duration": 60000
},
"citations": [
{
"url": "https://example.com",
"title": "Article Title",
"snippet": "Relevant excerpt...",
"author": "Author Name",
"publishedDate": "2024-01-01T00:00:00Z"
}
],
"data": {}
}
`

Development

`bash
bun install # Install dependencies
bun run build # Build project
bun run dev # CLI development
bun run dev:mcp # MCP server development
``

Performance

- ⚡ 885,371 requests/second - Search throughput
- 💾 85% cache hit rate - 40% API call reduction
- 🪶 <1MB footprint - Minimal memory overhead
- 🔄 1-20 parallel ops - Configurable concurrency

Security

- ✅ Comprehensive input validation
- 🔐 Secure API key handling
- 🔁 Exponential backoff with jitter
- 🛡️ Graceful degradation on failures
- ⏱️ Per-request timeout controls

---

License

MIT © 2026 - See LICENSE for details.

Links

- 🐛 Issues
- 💬 Discussions
- 📦 npm

Built With

- Exa - Neural search API
- Bun - JavaScript runtime
- MCP - AI integration
- TypeScript - Type safety

---