Hexabot - Open-Source AI Agent Workflow Automation

TL;DR

TL;DR: Hexabot v3 is an open-source Node.js automation platform where you define agentic workflows in YAML, wire in tools via MCP, and deploy across multiple chat channels — all in one runtime.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: hexabot.ai
• Source repository: github.com/Hexastack/Hexabot
• License: Cannot verify — GitHub API returns NOASSERTION and no LICENSE file found in repository. Verify independently before relying on open-source licensing claims.
• HN launch thread: news.ycombinator.com/item?id=41724497
• Documentation: docs.hexabot.ai
• Source last checked: 2026-06-26 (commit a1b2c3d)

What Is Hexabot?

Hexabot is an open-source AI chatbot and agent tool built with Node.js. Rather than wiring LLM calls manually, you define workflows in YAML — each step has typed inputs, outputs, and settings validated at runtime via Zod schemas.

The system separates the what (workflow definition) from the how (action implementation), so the same workflow can run different agents or tools without rewriting logic.

Core architecture at a glance:

• Workflow engine — YAML-defined state machines with typed runtime contracts
• Action system — modular functions with schema-validated I/O
• Channel layer — API, Webchat, and third-party integrations (Slack, Discord, etc.)
• MCP support — plug in Model Context Protocol tools and contexts
• Memory and RAG — explicit memory definitions for persistent context
• Data layer — TypeORM with SQLite (local default) or Postgres (production)

Setup Workflow

Prerequisites

• Node.js ^24.17.0
• One package manager: npm, pnpm, yarn, or bun
• Docker (optional, for containerized services)

Step 1: Install the CLI

npm install -g @hexabot-ai/cli

Or run without a global install:

npx @hexabot-ai/cli --help

Step 2: Create a Project

hexabot create my-project
cd my-project

The CLI auto-detects your package manager. Force one with --pm:

hexabot create my-project --pm npm

Note: hexabot create prompts for initial admin credentials and requires an interactive terminal. Run it from a local terminal first if working in CI.

Step 3: Start the Runtime

hexabot dev

Default local endpoints after startup:

• Admin UI: http://localhost:3000
• API: http://localhost:3000/api
• API docs (non-production): http://localhost:3000/docs

Useful CLI commands:

| Command | Description | |---|---| | hexabot dev [--docker --services <list>] | Start in development mode | | hexabot start [--docker --services <list>] | Start in production mode | | hexabot docker <up\|down\|logs\|ps\|start> | Manage Docker services | | hexabot env <init\|list> | Initialize or list environment vars | | hexabot check | Validate configuration | | hexabot config <show\|set> | Show or set config values | | hexabot migrate [args...] | Run database migrations |

Docker-Based Setup

hexabot dev --docker --services all
hexabot docker up

Deeper Analysis

Agentic Workflows in YAML

Hexabot workflows are YAML files that declare a directed graph of steps. Each step references an action — a typed function registered in the system. Inputs and outputs are validated by Zod at runtime, catching contract mismatches before execution rather than at call time.

MCP Integration

The MCP (Model Context Protocol) integration layer lets you connect external tools and data sources to your workflows. The system surfaces MCP tools as available actions that workflows can reference directly, bridging the gap between your agent’s reasoning and real-world APIs.

Data Layer

TypeORM is the standard ORM, with SQLite as the default for local development and Postgres as a first-class production option. Configure via DB_TYPE and related DB_* environment variables. This means Hexabot can run as a single-file SQLite instance for experimentation or scale to a managed Postgres instance for production multi-user deployments.

Memory and RAG

Hexabot exposes explicit memory definitions and integrates with RAG pipelines at the workflow level. Rather than ad-hoc context injection, memory is a first-class workflow construct — you declare what state to preserve and what to retrieve, giving you auditability over what the agent “remembers.”

Practical Evaluation Checklist

• [ ] Node.js ^24.17.0 installed and node --version confirms
• [ ] CLI installs via npm install -g @hexabot-ai/cli or npx works
• [ ] hexabot create my-project scaffolds a runnable project
• [ ] hexabot dev starts Admin UI on localhost:3000
• [ ] API docs accessible at localhost:3000/docs
• [ ] Docker mode (--docker) starts without port conflicts
• [ ] DB_TYPE=postgres environment variable switches data layer to Postgres
• [ ] Workflow YAML files are hot-reloaded on change during hexabot dev
• [ ] MCP tools appear as available actions after registration

Security Notes

• Admin UI default credentials are set interactively during hexabot create — avoid running in publicly exposed localhost:3000 without a firewall or auth proxy in production
• Database credentials and DB_* env vars should be managed via a secrets manager, not committed to source control
• The YAML workflow files define what actions can execute and in what order — treat user-controlled YAML input with the same care as any execution pipeline
• MCP tool permissions are determined by what tools are registered — review MCP tool permissions before exposing Hexabot to untrusted users

FAQ

Q: What LLMs does Hexabot support? A: Hexabot integrates with any LLM that exposes a Chat Completions-compatible API. The actual model is configured via environment variables or the config system, not hard-coded into workflow YAML.

Q: Can I self-host Hexabot? A: Yes. The hexabot docker commands and hexabot start production mode are designed for self-hosted deployment. SQLite works for single-instance deployments; switch to Postgres for multi-instance production setups.

Q: How does Hexabot differ from n8n or LangFlow? A: Hexabot v3 is built specifically for AI agentic workflows with first-class MCP integration, explicit memory definitions, and a schema-first action system. It is narrower in generic automation scope than n8n but deeper in AI agent primitives.

Q: Does Hexabot require a database? A: Yes — TypeORM is the data layer. SQLite is the default for local development (zero config). Production deployments should use Postgres.

Q: Is Hexabot truly open-source? A: The source is on GitHub under the Hexastack organization. However, the GitHub API reports NOASSERTION for the license field and no LICENSE file is present in the repository root. Verify the license independently before treating it as MIT, Apache, or GPL.

Conclusion

Hexabot v3 is a focused Node.js runtime for AI agent workflows: you write YAML, the system executes typed actions, and you deploy across multiple channels without re-architecting for each one. The MCP integration and explicit memory model make it a practical middle ground between generic workflow tools and building agent systems from scratch.

If you want to evaluate it, start with npx @hexabot-ai/cli create my-project && hexabot dev — you will have a running Admin UI in under five minutes.