At a Glance:
Hexabot v3 is an open-source automation platform with first-class AI capabilities that combines YAML-defined agentic workflows, action-based execution, and multi-channel conversational support with MCP integration, memory, and RAG.
Overview:
Hexabot v3 is an automation platform designed to build and run agentic workflows across conversational channels. It combines YAML-based workflow definitions, action-based execution with schema-validated inputs, and first-class AI capabilities including memory support, Model Context Protocol (MCP) integration, and retrieval-augmented generation (RAG). The platform provides a binding system that separates reusable capabilities from task logic, and supports multi-channel continuity where channels and helpers function as core concepts. Its schema-first architecture uses Zod for validation and shared contracts across the runtime. For data storage, Hexabot v3 uses TypeORM with SQLite as the default local option and first-class PostgreSQL support for production environments, configurable through database runtime variables.
Key Decision Points:
YAML workflow definitions: Agentic workflows are defined in YAML with typed runtime contracts, making workflow configuration explicit and version-controllable.
Schema-validated execution: Actions use Zod schemas for input, output, and settings validation, ensuring runtime type safety and clear contracts between workflow components.
MCP and memory integration: The platform explicitly supports Model Context Protocol for tool and context interoperability, alongside dedicated memory definitions for persistent state across conversations.
Database portability: TypeORM provides a consistent data layer, with SQLite for local development and PostgreSQL as the first-class production option, configurable via environment variables.
Multi-channel architecture: Channels and helpers are core architectural primitives, suggesting the platform is designed for deployment across multiple conversational interfaces rather than single-channel use.
Core Features:
Agentic workflow engine: Define automation workflows in YAML format with typed runtime contracts that govern execution behavior.
Action-based execution system: Workflow steps are implemented as actions with schema-validated inputs, outputs, and settings enforced through Zod.
Binding system: Reusable capability and configuration bindings are separated from task logic, promoting modular workflow composition.
Explicit memory support: Memory definitions are integrated into the runtime, allowing workflows and agents to maintain state across conversational turns.
Model Context Protocol (MCP) integration: MCP support provides interoperability for tools and context sharing between the platform and external AI systems.
Multi-channel continuity: The platform treats channels and helpers as core concepts, supporting conversation flow continuity across different communication channels.
Use Cases:
Developers building multi-channel conversational agents who need schema-validated, YAML-defined workflows with built-in memory and context management.
System architects designing automation systems that require separation of capabilities from task logic through a binding system and typed runtime contracts.
Teams prototyping agentic workflows locally with SQLite before deploying to production with PostgreSQL, using environment-based database configuration.
AI application builders seeking MCP-compatible runtime integration for tool use and context sharing across agentic workflows.
Open-Source Alternative Value:
Hexabot v3 provides developers with a schema-first, open-source automation runtime that integrates agentic workflows with MCP, memory, and RAG capabilities. Its YAML-based workflow definitions and Zod-validated action contracts offer explicit, type-safe configuration that can be version-controlled and reviewed. The binding system's separation of capabilities from task logic enables modular, reusable workflow components. With TypeORM providing a consistent data layer across SQLite and PostgreSQL, users can develop locally and transition to production without changing data access patterns.
