Skip to main content

​
Introduction

Welcome to BindAI. BindAI is an open-source Python framework for building production-ready AI applications with agents, workflows, tools, memory, knowledge retrieval (RAG), model providers, projects, external integrations, and multi-agent systems. Instead of building AI applications from scratch each time, BindAI provides a modular architecture that lets you compose intelligent systems from reusable building blocks. Whether you’re building a simple AI assistant, a tool-calling application, a knowledge-powered system, an automated workflow, or a multi-agent platform, BindAI provides a foundation that can grow with your application.

​
Why BindAI?

Modern AI applications quickly become more than a single LLM call. A production-ready AI system often needs:
  • AI agents
  • External tools
  • Conversation memory
  • Knowledge retrieval (RAG)
  • Embeddings
  • Multiple retrieval strategies
  • Workflow orchestration
  • Human approval and tasks
  • Scheduling
  • External service integrations
  • Multiple language model providers
  • Multi-agent coordination
  • MCP tool integration
Without a clear architecture, these capabilities can become tightly coupled and difficult to maintain. BindAI separates these responsibilities into modular packages so individual components can evolve independently while still working together as a complete application framework.

​
Core Principles

​
Modular

BindAI is organized into independent packages with clearly defined interfaces. You can use individual capabilities independently or combine them into a complete AI application. The modular architecture covers areas such as:
  • Agents
  • Tools
  • Providers
  • Memory
  • Embeddings
  • Retrieval
  • Knowledge
  • Workflows
  • Projects
  • Connections
  • MCP
This allows applications to start small and expand as requirements grow.

​
Provider Agnostic

BindAI is not tied to a single AI provider. The provider abstraction allows applications to work with multiple supported model providers through a consistent interface. Current supported providers include:
  • OpenAI
  • Anthropic
  • Google Gemini
  • Groq
  • Ollama
  • OpenRouter
Additional providers such as Azure OpenAI and Mistral are planned. Application logic can remain largely independent of the underlying provider, allowing providers to be changed as application requirements evolve.

​
Production Focused

BindAI is designed around the requirements of real-world AI applications. Current capabilities include:
  • Tool calling
  • Multiple tools
  • Structured agent configuration
  • Conversation management
  • Memory
  • Knowledge retrieval (RAG)
  • Embeddings
  • Vector retrieval
  • BM25 retrieval
  • Hybrid retrieval
  • Reranking
  • Workflow orchestration
  • Conditional execution
  • Loops
  • Parallel execution
  • Retry policies
  • Timeout handling
  • Scheduling
  • Human approval and tasks
  • Agent delegation
  • Multi-agent team execution
  • External service connections
  • MCP tool integration
  • Event handling
  • Middleware and callbacks
Additional production infrastructure, including observability and deployment capabilities, remains under development.

​
Python First

BindAI embraces modern Python development. Applications are built using familiar Python classes, functions, interfaces, decorators, and type hints. There is no custom scripting language or proprietary workflow syntax required to build BindAI applications.

​
Main Components

BindAI is composed of modular packages that work together.

​
Projects

Projects organize complete AI applications. A typical project can contain:
  • Agents
  • Workflows
  • Tools
  • Knowledge
  • Memory
  • Configuration
  • Templates
  • Integrations
Projects provide a consistent structure for organizing applications as they grow.

​
Agents

Agents are the primary execution abstraction for AI applications. Agents can:
  • Communicate with language models
  • Maintain conversations
  • Maintain application context
  • Call tools
  • Access memory
  • Retrieve knowledge
  • Produce structured results
  • Execute tool loops
  • Delegate work to other agents
  • Work with specialist agents
  • Participate in multi-agent teams
  • Use MCP-backed tools
Agents can also be extended with configuration, hooks, callbacks, middleware, retrieval, memory, and execution controls.

​
Providers

Providers connect BindAI agents to supported language model services. The provider layer provides a common interface for different model providers. Current providers include:
  • OpenAI
  • Anthropic
  • Google Gemini
  • Groq
  • Ollama
  • OpenRouter
The provider registry and bootstrap system allow supported providers to be discovered and initialized consistently. Additional providers are planned.

​
Tools

Tools allow agents to interact with functions and external systems. Tools can represent:
  • Python functions
  • Application services
  • REST APIs
  • Databases
  • Search systems
  • External integrations
  • MCP services
BindAI provides a tool registry and execution layer so agents can discover and execute tools through a consistent interface.

​
Memory

Memory allows applications and agents to store and retrieve information across interactions. Current memory capabilities include:
  • In-memory memory
  • SQLite
  • PostgreSQL
  • Vector memory
  • Pinecone
  • Chroma
  • Conversation memory
  • Custom memory providers
Memory implementations are designed behind common abstractions so applications can change storage strategies without tightly coupling agents to a particular backend.

​
Embeddings and Retrieval

BindAI provides embedding and retrieval abstractions for semantic and lexical search. Current capabilities include:
  • Embedding providers
  • Random/local embeddings
  • OpenAI embeddings
  • Vector retrieval
  • BM25 retrieval
  • Hybrid retrieval
  • Retrieval configuration
  • Search options
These capabilities form the foundation for knowledge retrieval and RAG applications.

​
Knowledge (RAG)

Knowledge provides Retrieval-Augmented Generation (RAG) capabilities. BindAI knowledge systems can:
  • Load documents
  • Ingest documents
  • Parse content
  • Split content into chunks
  • Generate embeddings
  • Store metadata
  • Perform semantic retrieval
  • Perform BM25 retrieval
  • Perform hybrid retrieval
  • Filter results
  • Rerank results
  • Perform conversational retrieval
  • Run knowledge pipelines
  • Supply retrieved context to agents
Knowledge can therefore be used independently or integrated directly into agent execution.

​
Workflows

The workflow engine orchestrates complex multi-step execution logic. Supported workflow patterns include:
  • Sequential execution
  • Conditional branching
  • Loops
  • Parallel execution
  • Retry policies
  • Timeout handling
  • Scheduling
  • Human approval
  • Human tasks
These capabilities can be combined to build reliable AI workflows and business automation processes.

​
Connections

Connections provide a consistent abstraction for integrating BindAI applications with external services. Current integrations include:
  • Webhooks
  • GitHub
  • Slack
  • Notion
  • Jira
  • Discord
  • Resend
  • Vercel
  • Netlify
The Connections package includes a connection abstraction, registry, and manager for handling connection lifecycle and service-specific implementations.

​
MCP

BindAI includes support for the Model Context Protocol (MCP). Current MCP capabilities include:
  • MCP client connections
  • Tool discovery
  • Tool calling
  • MCP tools exposed through the BindAI tool system
  • Basic connection handling
MCP support allows BindAI agents to work with tools provided by MCP-compatible services. Additional MCP protocol capabilities are planned.

​
Multi-Agent Systems

BindAI supports multiple agent execution patterns. Current capabilities include:
  • Agent delegation
  • Agent handoff
  • Specialist agents
  • Team delegation
  • Specialist role chains
  • RAG-enabled agents
  • Structured agent configuration
  • Agent execution configuration
These capabilities provide a foundation for applications where different agents perform specialized responsibilities. Advanced planning, supervisor agents, hierarchical coordination, and more advanced shared state management remain under development.

​
Command Line Interface

BindAI includes a CLI for working with BindAI projects and applications. The CLI provides the entry point for BindAI tooling and initializes the provider ecosystem before launching the application interface. The exact CLI capabilities evolve alongside the framework.

​
Typical Architecture

A typical BindAI application can combine several layers:
Applications can also introduce workflows, connections, MCP services, and additional agents:
Each layer has a clear responsibility, making applications easier to test, extend, and maintain.

​
Documentation Roadmap

If you’re new to BindAI, we recommend following this learning path:
  1. Installation
  2. Quick Start
  3. Your First Agent
  4. Core Concepts
  5. Providers
  6. Tools
  7. Memory
  8. Knowledge and RAG
  9. Workflows
  10. Projects
  11. Connections
  12. MCP
  13. Templates
  14. API Reference
Each guide builds on the previous one, gradually introducing the concepts needed to build increasingly capable AI applications. For the current implementation status and future development plans, see the ROADMAP.