Hexabot is an open-source AI automation platform designed for teams and developers who need to build and run agentic workflows across multiple channels like chat, email, and APIs. It solves the fragmentation problem in AI automation by unifying workflow definition, AI agent execution, memory management, and integration points into a single runtime. Unlike generic chatbot frameworks, Hexabot treats workflows as first-class entities with typed contracts, enabling complex, stateful AI automations that persist across interactions.
Built with TypeScript and Node.js, Hexabot uses TypeORM for data persistence (SQLite for local, PostgreSQL for production), Zod for schema validation, and supports Docker-based deployment. It integrates with major LLMs like GPT, Claude, Gemini, Llama, Mistral, and DeepSeek via MCP (Model Context Protocol), and allows custom actions and channels to be developed and extended. The platform is self-hosted, giving full control over data and infrastructure.
What You Get
- Agentic Workflows - YAML-defined workflows with typed runtime contracts that orchestrate AI agents, actions, and memory across channels.
- Action-based Execution - Reusable, schema-validated actions with defined inputs, outputs, and settings that encapsulate business logic and external API calls.
- MCP Integration - Native support for Model Context Protocol to connect external tools, plugins, and AI models (like Ollama, DeepSeek, Claude) directly into workflow actions.
- Memory System - Explicit memory definitions and runtime memory integration to maintain context across workflow runs and user interactions.
- Multi-channel Support - Built-in channels and helpers to handle conversations from Slack, Discord, email, webhooks, and custom sources with continuity.
- Schema-first Architecture - Comprehensive use of Zod for validation across workflows, actions, and bindings, ensuring type safety and reducing runtime errors.
Common Use Cases
- Running multi-step customer support automations - A SaaS company uses Hexabot to route inbound support tickets via email to an AI agent that retrieves knowledge base articles, checks user subscription status, and replies with personalized solutions.
- Automating internal HR onboarding - An HR team builds a workflow that triggers when a new hire is added to HRIS, auto-creates Slack channels, sends welcome emails, schedules training sessions, and updates Notion docs—all through a single YAML workflow.
- Building AI-powered research assistants - A research team deploys a Hexabot workflow that ingests PDFs, uses RAG to extract insights, and answers questions via a web chat interface powered by Llama 3 and Mistral.
- Orchestrating cross-platform data syncs - A developer creates a scheduled workflow that pulls data from Google Sheets, processes it with a custom action, updates a PostgreSQL table, and notifies a Discord channel—all with error handling and retry logic.
Under The Hood
Architecture
- The project employs a monorepo structure, fostering code sharing and optimized builds across distinct packages.
- A layered architecture is evident, with packages dedicated to specific concerns like agentics, API, data types, UI, and CLI tools.
- The structure promotes modularity, but a deeper dive is needed to fully assess coupling and SOLID principle adherence.
Tech Stack
- The core logic is built using TypeScript and the NestJS framework, indicating a focus on scalability and maintainability.
- TypeORM is utilized for database interaction, suggesting a relational database backend.
- A robust testing strategy is in place, leveraging Jest and comprehensive linting tools.
- Dependency management is proactive, with overrides used to resolve compatibility issues.
Code Quality
- Comprehensive testing is evident, including unit and integration tests with a focus on schema validation and service behavior.
- Code organization is generally well-structured, with clear separation between API and frontend components.
- Error handling is implemented with custom error classes and structured logging.
- Consistent naming conventions and TypeScript’s type safety enhance readability and maintainability.
What Makes It Unique
- Specialized AI SDKs are integrated, suggesting a sophisticated approach to AI-driven chatbot functionality.
- The use of jsonata for data transformation within the agentic package is a distinctive feature.
- A custom schema validation approach and a CLS (Contextual Logging System) demonstrate a focus on observability and security.
- The shared output contract system using runtime entity schemas promotes type safety and interoperability.