Skip to main content
Version: 1.0

Integrations

What are Integrations?​

Integrations are special assets that connect external services to aiXplain, generating tools that you can use in your agents. When you connect an integration, it creates a private tool with all the capabilities of that external service.

Key characteristics:

  • All integrations are currently public (visible to all aiXplain users)
  • Connecting an integration generates a private tool (visible only to your team)
  • You can create multiple tools from the same integration
  • Generated tools can be used in any of your agents

Finding Integrations​

  1. Navigate to Browse in Studio
  2. Click on the Integrations tab
  3. Browse available integrations or use the search bar
  4. Click on any integration card to view details

Connecting an Integration​

General Connection Process​

  1. Click Connect from the integration's menu or detail page
  2. Provide a name for the tool you're creating
  3. Provide required authentication or configuration details
  4. Click Create or Connect
  5. A private tool is automatically generated and appears in your tools

Note: You can create multiple tools from the same integration with different configurations or credentials.

Standard Integrations​

Standard integrations connect to third-party APIs and services.

Connection Requirements​

When connecting a standard integration, you'll typically need to provide:

  • Tool name - What you want to call the generated tool
  • Authentication - Using one of these methods:
    • API Key - A secret key from the service provider
    • OAuth2 - Modern authorization protocol
    • OAuth1 - Legacy authorization protocol

How They Work​

Once connected:

  • The integration reads available actions from the external service
  • A private tool is created with all those actions
  • The tool description includes information about what the service does
  • Your agents can use this tool like any other tool

Premium Integrations​

aiXplain offers four specialized premium integrations designed for common enterprise use cases.

SQLite Integration​

Developer: aiXplain

Connect a SQLite database and interact with it using natural language through your agents.

Connection Requirements​

  • Tool name - Name for your database tool
  • SQLite database file - Upload your .sqlite or .db file

Current limitation: The database must contain only one table.

How It Works​

  1. Upload your SQLite database file
  2. The schema is automatically read
  3. The schema is added to the tool's description
  4. Agents can see the schema and understand how to query your database
  5. Agents can construct and execute SQL queries based on natural language requests

Available Actions​

The generated tool provides two actions:

  • Query - Read data from the database

    • Execute SELECT statements
    • Retrieve records based on conditions
    • Join tables (when multi-table support is added)

  • Commit - Write data to the database

    • Execute INSERT, UPDATE, DELETE statements
    • Modify database records
    • Create new entries

Use Cases​

  • Customer data lookup for support agents
  • Product inventory queries
  • Order status checking
  • Internal knowledge base access
  • Data analysis and reporting

Example​

User to Agent: "How many customers do we have in California?"
Agent uses SQLite tool to query: SELECT COUNT(*) FROM customers WHERE state = 'CA'
Agent responds: "You have 1,247 customers in California."

Python Sandbox Integration​

Developer: aiXplain

Execute custom Python scripts securely in an isolated sandbox environment.

Connection Requirements​

  • Tool name - Name for your Python tool
  • Python file - Upload your .py file
  • Main function name - Specify which function is the entry point

Note: Your Python file can contain multiple functions, but you must specify which one serves as the main entry point.

Python File Requirements​

Your Python script must follow a specific signature format. See the technical documentation for detailed requirements.

General guidelines:

  • Define a clear main function
  • Accept parameters that agents can pass
  • Return results in a structured format
  • Handle errors gracefully

How It Works​

  1. Upload your Python script
  2. Specify the main function
  3. The tool is created with your script
  4. When an agent calls the tool, your script executes in a secure sandbox
  5. Results are returned to the agent

Use Cases​

  • Custom data transformations
  • Complex calculations and algorithms
  • API interactions not covered by standard integrations
  • Data validation and processing
  • Custom business logic execution

Example​

def analyze_sentiment(text):
    # Custom sentiment analysis logic
    score = calculate_sentiment_score(text)
    return {
        "sentiment": "positive" if score > 0.5 else "negative",
        "confidence": score
    }

MCP Server Integration​

Connect any remote MCP (Model Context Protocol) server that supports API key authentication.

What is MCP?​

Model Context Protocol (MCP) is a standardized protocol for connecting AI systems to external data sources and tools. MCP servers expose capabilities that AI agents can discover and use.

Connection Requirements​

  • Tool name - Name for your MCP tool
  • MCP Server URL - The endpoint of your MCP server
  • API Key - Authentication credentials for the server

How It Works​

  1. Provide your MCP server URL and API key
  2. aiXplain connects to the server and discovers available actions
  3. A private tool is generated with all actions the MCP server supports
  4. Your agents can immediately use these actions
  5. Requests are routed through the MCP protocol to your server

Generated Tool​

The tool automatically includes:

  • All actions exposed by your MCP server
  • Action descriptions from the server
  • Input/output schemas for each action
  • Authentication handling

Use Cases​

  • Connect to enterprise systems with MCP support
  • Access proprietary data sources
  • Integrate custom internal tools
  • Bridge legacy systems to AI agents
  • Create reusable integrations across your organization

AIR Vector Database Integration​

AIR = AI Information Retrieval

Connect to aiXplain's vector database engine for semantic search, document storage, and retrieval.

What is a Vector Database?​

Vector databases store information as mathematical vectors (embeddings), enabling semantic search. Unlike traditional keyword search, semantic search understands meaning and context.

Example:

  • Traditional search for "feline" won't find documents about "cats"
  • Semantic search understands they're related concepts

Connection Requirements​

  • Vector database name - Name for your database
  • Embedding model - Choose which model converts text to vectors
    • Default: OpenAI's embedding model
    • Options: Any embedding model available on the platform
    • Custom: Request aiXplain to onboard additional models

Setting Up Your Vector Database​

  1. Connect the integration with a name and embedding model
  2. A private tool is generated
  3. Upload your content:
    • Images
    • PDFs
    • Text files
    • Additional supported formats (shown in the connection interface)

Note: All uploaded files are converted to vectors using your selected embedding model.

Available Actions​

The AIR tool provides six actions:

Search

  • Perform semantic search across stored content
  • Query with natural language
  • Returns most relevant results based on meaning, not just keywords
  • Includes similarity scores

Count

  • Get the total number of records in the database
  • Useful for monitoring database size
  • No parameters required

Upsert

  • Add new records to the vector database
  • Update existing records
  • Automatically generates embeddings
  • Can be called by agents to store information

Get

  • Retrieve a specific record by ID
  • Returns the full record content
  • Useful for fetching known items

Delete

  • Remove a specific record by ID
  • Permanently deletes from the database
  • Cannot be undone

Retrieve Metadata

  • Get metadata for a specific record
  • Returns information about the record without full content
  • Useful for managing database entries

Use Cases​

Retrieval-Augmented Generation (RAG):

  • Store company documentation
  • Enable agents to search and cite sources
  • Provide context for accurate responses

Knowledge Management:

  • Build searchable knowledge bases
  • Store and retrieve product information
  • Create internal wikis accessible to agents

Document Search:

  • Semantic search across company documents
  • Find relevant policies or procedures
  • Search technical documentation

Customer Support:

  • Store support articles and FAQs
  • Enable agents to find relevant solutions
  • Provide accurate, sourced answers

Research and Analysis:

  • Store research papers and reports
  • Enable semantic literature review
  • Connect findings across documents

Example Workflow​

1. User uploads 100 company policy documents to AIR
2. Agent receives question: "What's our remote work policy?"
3. Agent uses AIR Search action with query: "remote work policy"
4. AIR returns top 3 relevant document sections
5. Agent reads the content and provides an answer with citations

Managing Generated Tools​

All tools created from integrations are private assets that belong to your team.

What You Can Do​

Edit Metadata

  • Modify the tool's name
  • Update the description to provide better context for agents
  • Changes only affect your tool, not the integration

Use in Agents

  • Add to any single agent
  • Add to team agents
  • Specify which actions to include

Transfer

  • Move the tool to another team you belong to
  • The tool is completely transferred (not copied)
  • You lose access in the original team

Delete

  • Permanently remove the tool
  • Cannot be undone
  • Agents using this tool will no longer have access

Best Practices​

  • Name tools descriptively - "Customer Database" is better than "SQLite Tool 1"
  • Write clear descriptions - Help agents understand when and how to use the tool
  • Test before adding to agents - Use the Tryout tab if available
  • Document credentials - Keep track of API keys and authentication separately
  • Review permissions - Ensure the tool has appropriate access levels

Integration Limitations and Considerations​

Third-Party Dependencies​

Important: Integrations rely on third-party providers that control their own systems.

Potential issues:

  • Rate limits - Provider may limit requests per minute/hour
  • Quota limits - Provider may cap total usage
  • Downtime - Provider service may be temporarily unavailable
  • API changes - Provider may update or deprecate endpoints

aiXplain's responsibility:

  • We provide the connection infrastructure
  • We do not control third-party availability or performance
  • We do not assume liability for provider limits or errors

Authentication Security​

  • Store credentials securely - Don't share API keys unnecessarily
  • Rotate keys regularly - Update authentication credentials periodically
  • Use minimum permissions - Grant tools only the access they need
  • Monitor usage - Watch for unexpected activity

Cost Considerations​

  • Provider costs - Some services charge per API call
  • aiXplain credits - Tool execution consumes credits
  • Embedding costs - AIR uses embedding models (charged separately)
  • Storage costs - Large vector databases may have storage implications

Troubleshooting​

Connection Fails​

Common causes:

  • Invalid credentials
  • Incorrect URL format
  • Firewall blocking connection
  • Service temporarily unavailable

Solutions:

  1. Verify credentials with the service provider
  2. Check URL format (include https://)
  3. Test credentials directly with the service
  4. Try again after a few minutes

Tool Doesn't Work in Agent​

Common causes:

  • Tool actions not properly selected
  • Agent lacks proper instructions
  • Authentication expired
  • Service rate limit reached

Solutions:

  1. Verify which actions are enabled
  2. Update agent instructions to describe when to use the tool
  3. Reconnect the integration with fresh credentials
  4. Wait for rate limit to reset (check provider documentation)

Search Results Not Relevant (AIR)​

Common causes:

  • Wrong embedding model for your content type
  • Insufficient uploaded content
  • Query doesn't match content style

Solutions:

  1. Try a different embedding model
  2. Upload more comprehensive content
  3. Rephrase queries to better match document language
  4. Add more context to search queries