airweave-setup（爱维维设置）

Airweave Setup & Integration

Airweave is an open-source platform that makes any app searchable for AI agents. It connects to apps, productivity tools, databases, or document stores and transforms their contents into searchable knowledge bases.

Quick Start

Option 1: Airweave Cloud (Recommended)

1. Sign up at https://app.airweave.ai
2. Get your API key from the dashboard
3. Install the SDK:

pip install airweave-sdk   # Python
npm install @airweave/sdk  # TypeScript

Option 2: Self-Hosted

git clone https://github.com/airweave-ai/airweave.git
cd airweave
chmod +x start.sh
./start.sh

Access the dashboard at http://localhost:8080

Core Workflow

The typical Airweave workflow follows these steps:

1. Create a Collection

A collection groups multiple data sources into a single searchable endpoint.

Python:

from airweave import AirweaveSDK

client = AirweaveSDK(
    api_key="YOUR_API_KEY",
    base_url="https://api.airweave.ai"  # or http://localhost:8001 for self-hosted
)

collection = client.collections.create(name="My Knowledge Base")
print(f"Collection ID: {collection.readable_id}")

TypeScript:

import { AirweaveSDKClient } from "@airweave/sdk";

const client = new AirweaveSDKClient({ apiKey: "YOUR_API_KEY" });
const collection = await client.collections.create({ name: "My Knowledge Base" });

2. Add Source Connections

Connect data sources to your collection. 40+ sources supported including:

• Productivity: Notion, Google Drive/Docs/Slides, Dropbox, OneDrive, SharePoint, Box, Airtable
• Communication: Slack, Gmail, Outlook, Teams, Google Calendar
• Project Management: Jira, Linear, Asana, Trello, Monday, ClickUp, Todoist
• Development: GitHub, GitLab, Bitbucket, Confluence
• CRM & Sales: Salesforce, HubSpot, Attio, Zendesk, Pipedrive, Shopify
• Data: Stripe, PostgreSQL

See SDK-REFERENCE.md for the complete list of source short names.

Python (API Key sources like Stripe, Linear):

source = client.source_connections.create(
    name="My Stripe Connection",
    short_name="stripe",
    readable_collection_id=collection.readable_id,
    authentication={
        "credentials": {"api_key": "sk_live_your_stripe_key"}
    }
)

OAuth sources (Slack, Google, Microsoft, etc.):

Most sources use OAuth for authentication. Use the Airweave UI at https://app.airweave.ai to connect these sources—it handles the OAuth flow automatically.

3. Search Your Data

Once synced, search across all connected sources. Airweave provides three search modes:

Python:

# Classic search (default) — AI-optimized
results = client.collections.search(
    readable_id=collection.readable_id,
    query="customer feedback about pricing",
    mode="classic"
)

for result in results.results:
    print(f"Source: {result.airweave_system_metadata.source_name}")
    print(f"Name: {result.name}")
    print(f"Content: {result.textual_representation[:200]}...")
    print(f"Score: {result.relevance_score}")
    print(f"URL: {result.web_url}")

Search Modes in Detail

Instant Search

Fast, direct vector search. Best for simple lookups.

results = client.collections.search(
    readable_id=collection.readable_id,
    query="API documentation",
    mode="instant",
    retrieval_strategy="hybrid",  # "hybrid" (default), "neural", or "keyword"
    limit=20,
    offset=0
)

Classic Search

AI generates an optimized search plan. Best general-purpose mode.

results = client.collections.search(
    readable_id=collection.readable_id,
    query="technical documentation",
    mode="classic",
    limit=20,
    offset=0
)

Agentic Search

Full agent loop that iteratively searches, reads, and reasons over results.

results = client.collections.search(
    readable_id=collection.readable_id,
    query="Summarize all decisions about the authentication redesign",
    mode="agentic",
    thinking=True,  # Enable extended reasoning
    limit=10
)

MCP Integration for AI Agents

Airweave exposes search via MCP (Model Context Protocol) for seamless AI agent integration.

Setup for Claude Desktop / Cursor

Add to your MCP configuration (~/.cursor/mcp.json):

{
  "mcpServers": {
    "airweave-search": {
      "command": "npx",
      "args": ["airweave-mcp-search"],
      "env": {
        "AIRWEAVE_API_KEY": "your-api-key",
        "AIRWEAVE_COLLECTION": "your-collection-id",
        "AIRWEAVE_BASE_URL": "https://api.airweave.ai"
      }
    }
  }
}

Hosted MCP Server

For cloud-based AI platforms like OpenAI Agent Builder:

• URL: https://mcp.airweave.ai
• Uses Streamable HTTP transport (MCP 2025-03-26)

See MCP-SETUP.md for detailed configuration.

Troubleshooting

No Results Found

• Check that sync has completed (can take a few minutes for large sources)
• Verify the collection ID is correct
• Try a broader search query
• Try a different search mode (e.g., classic instead of instant)

Authentication Errors

• Verify your API key is valid
• Check that the API key has access to the collection
• For OAuth sources, the token may have expired—reconnect in the UI

Rate Limits

• The API has rate limits for protection
• Implement exponential backoff for retries
• Contact support for higher limits

Additional Resources

• Documentation: https://docs.airweave.ai
• API Reference: https://api.airweave.ai/docs (Swagger)
• GitHub: https://github.com/airweave-ai/airweave
• Discord: https://discord.gg/484HY9Ehxt

For detailed SDK reference, see SDK-REFERENCE.md. For advanced search patterns, see SEARCH-PATTERNS.md.