Skip to main content
ToolKit Architecture

ToolKit provides your AI agents with intelligent, authenticated access to 150+ integrations

ToolKit is the complete infrastructure and data layer for building AI agents that interact with third-party integrations. It provides a single set of tools that work across all major AI frameworks—Vercel AI SDK, LangChain, OpenAI Agents, Mastra, and MCP—giving your agents the intelligence to execute actions flawlessly.
ToolKit powers production AI agents processing thousands of integration actions daily. Create a free account to give your agent access to 150+ integrations in minutes.

Why use ToolKit?

Building AI agents that reliably interact with third-party APIs is complex. ToolKit solves this by providing:

Intelligence built-in

Pica’s knowledge base handles schemas, edge cases, dependencies, and best practices for every action across 150+ integrations

Enterprise security

Automatic authentication, token management, and compliance built-in—no credential storage in your codebase

Framework agnostic

Works with any AI framework or LLM—Vercel AI SDK, LangChain, OpenAI, Claude, Mastra, MCP, and more

Fine-grained control

Scope access by connection, action, and permission level to control exactly what your agent can do

How it works

ToolKit sits between your AI agent and third-party integrations, providing intelligent tool execution with any LLM:

The execution flow explained

1

User makes a request

Your user asks the AI agent to perform an action in natural language (e.g., “Send a summary email to the team” or “Create a lead in Salesforce”).
2

Agent selects tools

Your AI agent (powered by any LLM—GPT-4, Claude, Gemini, etc.) selects the appropriate ToolKit tools based on the request.
3

ToolKit validates with knowledge

ToolKit consults Pica’s knowledge base to validate request schemas, handle dependencies, and apply best practices for that specific integration.
4

Authenticated execution

ToolKit makes the authenticated API request using securely stored credentials, handling rate limits, retries, and edge cases automatically.
5

Result returned to agent

The formatted response is returned to your agent, which synthesizes it into a natural language response for the user.
ToolKit works with any LLM and any framework. Whether you’re using OpenAI, Anthropic, Google, or open-source models, ToolKit provides the same intelligent tool execution layer.

What makes ToolKit powerful?

Pica’s Knowledge Base

The secret to ToolKit’s reliability is Pica’s comprehensive knowledge base—a constantly evolving intelligence layer that ensures flawless action execution. What the knowledge base covers:
Complete request and response schemas for 25,000+ actions across 150+ integrations. Your agent always sends properly structured data.
Rate limits, pagination, nested objects, required field combinations, API quirks, and error recovery patterns—all handled automatically.
Understands when actions require specific data from other endpoints, what fields depend on each other, and the proper sequence of operations.
Pre-built workflows that combine multiple actions intelligently (e.g., “create contact, add to list, send email” as a single operation).
Field formatting preferences, optimal parameter combinations, and integration-specific conventions for reliable results.
The knowledge base evolves with API changes, new endpoints, and real-world usage patterns—keeping your agent ahead of the curve.
This knowledge base is what transforms ToolKit from a simple API wrapper into an intelligent execution layer. Your agent doesn’t need to learn API documentation—Pica’s knowledge ensures every action works perfectly from day one.

Fine-grained access control

ToolKit gives you precise control over what your agent can do:

Connection Scope

Limit access to specific connections. Each agent can only act on the connections you specify.

Action Scope

Restrict which operations or API actions the agent is allowed to perform.

Permission Scope

Control the types of operations (read, write, admin) the agent can execute.
This enables you to build secure, multi-tenant agents where each user’s agent only accesses their own connected accounts and performs authorized actions.

Use cases

ToolKit powers AI agents across every industry and workflow:
  • Compose and send Gmail emails containing meeting summaries to team members
  • Schedule Google Calendar events with specified date/time parameters
  • Post messages with campaign analytics to designated Slack channels
  • Search and retrieve Q3 planning materials from Google Drive
  • Manage Notion workspaces with project tracking databases
  • Execute PostgreSQL queries to identify top customer segments
  • Generate new Google Sheets workbooks populated with sales metrics
  • Retrieve closing opportunity data from Salesforce CRM
  • Analyze customer behavior patterns across multiple data sources
  • Create automated reports combining data from various integrations
  • Create customer support cases in Zendesk from feedback data
  • Process customer refund transactions via Stripe
  • Convert website inquiries into HubSpot lead entries
  • Generate client invoices through QuickBooks integration
  • Automate order fulfillment workflows across systems
  • Create DALL-E images matching product requirements
  • Convert meeting audio to text using ElevenLabs
  • Conduct market research via Tavily/SerpApi integrations
  • Perform sentiment analysis on support interactions
  • Generate and distribute content across multiple platforms
Want to see what you can build with ToolKit? Check out real-world examples in our Showcase.

Framework integrations

Choose your framework and get started in minutes:

What’s next?

I