Metabot Platform
Architecture Features and Functionality
Abstract
This document provides a comprehensive overview of the Metabot platform architecture, its current implementation state, and roadmap for development. It describes the platform’s structural layers, data model, plugin system, execution engine, and communication modules, as well as upcoming extensions toward multi-agent orchestration, cognitive layers, and generative AI integration. The paper is intended for architects, developers, and integrators seeking a clear understanding of Metabot’s technical foundation and evolution path.
Metabot is a universal platform for designing, executing, and integrating interaction scenarios between users and systems.
Its core was originally built as a tool for bot design, business process automation, messenger marketing, and data management, but architecturally, Metabot already serves as a foundation for building multi-agent systems, neuro-integrations, digital contact centers, and cognitive layers powered by Generative AI.
Core Architecture (Implemented)
Main Platform Entities
- Bot — a program integrated with messaging platforms (Telegram, WhatsApp, WebChat, etc.).
Each bot executes scenarios consisting of commands and JavaScript code. - Lead — a user interacting with the bot. All user data, actions, and communication context are stored in the system.
- Attributes, Personas, Tickets — predefined core entities upon which interaction logic is built.
🟢 These elements are already implemented and form the foundation of any business logic within Metabot.
Scenario and Command Architecture
The platform supports two levels of scenario control:
- System Commands with fixed logic ("Send message", "Request data", "Set tag", "Set variable", etc.) —
Low-code mechanisms for communication designers. - JS Commands — fully executable fragments of JavaScript code compiled and executed on the production server at runtime.
This allows instant updates to business logic without CI/CD deployment.
🟢 Both levels are implemented and actively used in scenario design.
Plugins and Extensibility
- The platform supports the V8 engine, enabling the execution of JavaScript and PHP code.
- It allows creation of business plugins (within a single business account) and shared plugins (available to all bots on the server).
- Plugins can invoke each other’s methods, forming a reusable library of ready-to-use solutions.
🟢 The base plugin mechanism is implemented.
🔵 Planned: expanding the plugin system into a full-fledged marketplace of shared solutions with version control, dependency tracking, and code signing.
Data Model and Table Designer
Metabot includes a visual data model designer that provides advanced capabilities:
- Table design — creation and configuration of tables with various data types
- Relationships between tables — visual construction of one-to-many, many-to-many, and one-to-one relations
- Automatic form generation — automatic creation of complex forms based on table relationships
- Cascading operations — control of behavior on deletion or update of related records
- Data validation — built-in and custom integrity checks
- Intelligent interfaces — forms automatically adapt to relationship type (dropdowns, multi-selects, autocomplete)
Advantages of the approach:
- Reduced development time through automatic UI generation
- Data integrity ensured via relational consistency
- Simplified handling of complex data structures
- Visual representation of the database structure
🟢 Core functionality is already implemented and widely used.
🔵 Planned improvements: data schema versioning, automatic model updates during artifact deployment, and enhanced ER-diagram visualization.
Messenger Marketing and Automation (Implemented)
The platform includes a set of ready-to-use messenger marketing tools, which are actively used in production:
- Broadcasts — mass and personalized messages sent through connected messaging platforms.
- Autofunnels — multi-step interaction scenarios with leads, including trigger-based messages, segmentation, and state transitions.
- Gamified Scenarios — interactive communication through quizzes, quests, points, and rewards, implemented via JS commands and bot states.
🟢 All of this functionality is implemented and actively used in projects.
Metadesk Contact Center (Implemented)
Metadesk is the built-in multichannel contact center within the platform, providing a seamless transition between bot automation and human operators.
Key Capabilities
- Unified Operator Workspace — enables handling clients across multiple bots simultaneously.
- Automatic Dialogue Distribution — load balancing with respect to operator status (online/offline) and custom routing logic via JS.
- Adaptive Interface — accessible from desktops and mobile devices without additional installations.
- Notifications — sound alerts (only when the window is minimized) and visual indicators for new messages and clients.
- Dialogue Grouping — tabs such as “Waiting”, “My Dialogues”, and others.
- Customization — logic and UI can be extended or modified via JavaScript.
🟢 Metadesk is fully implemented and integrated into the platform.
Website Assistant Widget (Implemented)
The platform provides a customizable assistant widget for integration into client websites.
Widget Components
- Branded Interface — design adapted to the client’s website style.
- Chat Interface — message history, input field, and generative responses.
- Additional Panels — slides, presentations, links to documents.
- “Talking Head” — animated avatar with text-to-speech (TTS) synthesis.
- Responsiveness — supports desktop, tablet, and mobile devices.
🟢 The widget is fully implemented and can be configured according to client requirements.
Cognitive Layers: Intelligent Meaning Storage (In Development)
Cognitive Layers represent a new architectural concept of the platform, aimed at creating intelligent storage systems and meaning aggregation mechanisms embedded into enterprise processes.
Key Components
- Semantic Storages — vector databases for storing and searching semantic patterns.
- Context Aggregators — systems for collecting and analyzing information from multiple sources.
- Meaning Processors — modules for extracting and interpreting meaning from data.
- Cognitive Interfaces — APIs for integrating intelligent capabilities into business workflows.
Advantages of the Approach
- Deep personalization of interactions based on semantic profiles.
- Proactive recommendations derived from analysis of semantic patterns.
- Adaptive scenarios — automatic logic adjustment based on context.
- Cognitive analytics — discovery of hidden correlations and trends.
🔵 Currently under active development: Cognitive Layer architecture, semantic processors, and integration with the multi-agent system.
Execution and Update Architecture (Partially Implemented)
Scenario Execution
All scenarios and commands are executed in the context of a Bot and a Lead.
Each process maintains its own state, which allows the system to:
- resume a session after interruption;
- manage branches and re-entries;
- log every action for full traceability.
🟢 The core state management and logging mechanism is implemented.
🔵 Planned: implementation of detailed scenario tracing (debug panel, step-by-step analysis, and a visual execution profiler).
Real-Time Scenario Updates (No Deployment Required)
Since JS code is compiled at runtime, logic updates can be made directly on the production server in real time.
This provides exceptional agility for marketers and analysts.
🟢 The mechanism is already functional.
🔵 Planned improvements: introducing scenario version control and a pre-testing system (A/B sandbox) to combine flexibility with operational safety.
Versioning, Testing, and Artifacts (Planned)
To ensure sustainable development of complex solutions, the platform is transitioning to an artifact package model —
containers that include:
- data model updates;
- plugins and JS modules;
- scenarios and business logic;
- API specifications and dependencies;
- cognitive models and semantic processors.
Each artifact will have:
- a unique identifier and metadata (version, date, author);
- rollback capability;
- a declared dependency manifest.
🔵 In development:
- Artifact and scenario versioning system
- A/B testing and staged rollout of updates
- Automated testing and CI validation of packages before publication
- Dependency management between plugins and data tables
- Versioning of cognitive models and semantic processors
Multi-Agent Architecture and Generative AI (In Progress)
The platform is already being used as an environment for building multi-agent systems,
where each agent can be linked to a specific AI model:
- Supported integrations with OpenAI, Google, Anthropic, Meta, DeepSeek, and others via API.
- Work is underway on embedding local models (Llama, Gemma, Qwen) — both within client infrastructure and within the Metabot perimeter.
- Agents will have access to all internal functionalities: data, plugins, APIs, scenarios, and Cognitive Layers.
🟢 Partially implemented.
🔵 In active development: a unified agent management system, including contexts, roles, permissions, and orchestration of interactions between agents and bots, as well as integration with Cognitive Layers.
Analytics and Traceability (Planned Expansion)
- Execution logs and error monitoring are already implemented.
- Planned: creation of a visual scenario tracker, built-in analytics dashboards, and a performance metrics system for bots and agents.
- Cognitive analytics — analysis of semantic patterns and effectiveness of meaning processors.
🔵 The goal is to provide architects and marketers with tools to analyze conversion, response speed, and logic efficiency without relying on external BI tools — including monitoring of cognitive layers.
Solutions Marketplace (Planned)
To enable distribution of developed plugins, JS modules, artifacts, and cognitive models,
a built-in marketplace is planned, where:
- developers will be able to publish and sell their solutions;
- users will be able to install plugins, scenarios, and semantic processors with one click;
- the system will automatically track versions and dependencies.
🔵 This component is not yet implemented but is included in the architectural roadmap.
Low-code + Full-code Symbiosis (Implemented and Evolving)
Metabot combines visual tools for scenario design (Low-code) with full JS execution capabilities (Full-code).
This hybrid approach enables:
- Marketers and communication designers to build complex processes without developers.
- Developers to embed intelligent algorithms and integrations directly within scenarios.
- Cognitive architects to configure semantic processors and meaning aggregators.
🟢 Already implemented.
🔵 Planned: expansion of the visual editor into a full-fledged BPMN-style constructor with support for cognitive layers.
Low-code + Full-code Symbiosis (Implemented and Evolving)
Metabot combines visual tools for scenario design (Low-code) with full JS execution capabilities (Full-code).
This hybrid approach enables:
- Marketers and communication designers to build complex processes without developers.
- Developers to embed intelligent algorithms and integrations directly within scenarios.
- Cognitive architects to configure semantic processors and meaning aggregators.
🟢 Already implemented.
🔵 Planned: expansion of the visual editor into a full-fledged BPMN-style constructor with support for cognitive layers.
Evolution: Platform Development Vector
| Direction | Current State | Development Plan |
|---|---|---|
| JS Commands | Server-side compilation, instant logic updates | Add versioning and A/B sandbox |
| Plugins | JS/PHP plugins for bots and businesses | Centralized marketplace + dependency management |
| Data Model Designer | Visual table and relation design | Advanced ER-diagram visualization and migrations |
| Form Autogeneration | Basic forms from table relations | Advanced UI with customizable logic |
| Multi-Agent System | External model connections | Full agent orchestration system |
| Cognitive Layers | Concept and architectural design | Implementation of semantic storages and meaning processors |
| Testing | Partial | Automated tests and A/B testing infrastructure |
| Integrations | API, messengers, CRM supported | Visual API designer and flow editor |
| Analytics | Logs and tracing | Dashboards, scenario profiler, and cognitive analytics |
| Metadesk Contact Center | Fully implemented | Extended analytics and AI-agent integration |
| Website Assistant Widget | Implemented | Enhanced TTS and avatars based on client requests |
| Messenger Marketing | Implemented | Improved funnel analytics and gamification tools |
Conclusion
Metabot is a next-generation platform that combines a robust core for bot design, messenger marketing,
contact center management, and business logic with an ambitious roadmap toward multi-agent systems,
Cognitive Layers, a visual data model designer, artifact versioning, A/B testing, a solutions marketplace,
and generative AI integration.
Today, Metabot already unites:
- the speed of Low-code development,
- the flexibility of a Full-code environment,
- a powerful data model designer with automatic form generation,
- ready-made solutions for communication and automation,
- and enterprise-grade architectural stability.
In upcoming iterations, Metabot will evolve into a comprehensive constructor for neuro-agents, cognitive layers,
and digital business processes — a platform where logic, data, semantic models, and AI agents are all managed
as part of a unified artifact system.
Cognitive Layers and the visual data model designer will transform the platform into an intelligent ecosystem
capable not only of automating processes, but also of understanding meaning, adapting to context,
and proactively improving business interactions through deep semantic analysis and well-structured data architecture.
Part of the Next Paradigm Foundation research series on Communication Operating Systems.