Skip to content

Atlas Overview

The Open Source Framework for AI-Driven Community Platforms and Personal Agents

Overview

architecture.png

Atlas is an open-source framework that transforms how communities and individuals interact with AI agents, knowledge bases, and automated workflows. Built on battle-tested infrastructure (GURU Framework, Camunda BPMN, LangGraph), Atlas provides:

  • Dynamic RAG - Self-propelling knowledge bases from streamed data (Telegram groups, channels, documents)
  • Intercom-Style Assistant - Open-source support layer connecting web and Telegram
  • Agentic Workflows - Comprehensive scenarios with forms, tools, and business process automation
  • One-Click Deployment - Launcher wizard that spins up complete applications in Kubernetes namespaces

Atlas bridges the gap between simple chatbots and enterprise-grade automation, making sophisticated AI agents accessible to everyone.

1. Value Proposition

The Problem

Most AI assistants today are either:

  • Too simple - One-shot prompts with no memory, context, or business logic
  • Too complex - Enterprise platforms requiring months of integration and custom development
  • Too siloed - Separate systems for local, web, Telegram, knowledge bases, and workflows

The Solution

Atlas provides a complete system for building AI-driven applications that:

  1. Learn continuously from streamed data (messages, documents, events)
  2. Have comprehensive sub-agents, tools, and structure - Multi-agent systems with specialized capabilities
  3. Work across interfaces (web, Telegram, API) with shared context
  4. Execute business logic through rendered BPMN workflows and agentic scenarios
  5. Deploy in minutes via the launcher wizard, then customize from your fork

Core Value

Atlas is not just a chatbot framework - it's a complete orchestration platform where:

  • Knowledge bases are dynamic (self-updating from live sources)
  • Agents have personas, skills, and contexts (not just prompts)
  • Workflows are comprehensive (driving users through pre-set scenarios with conditions, forks, and tools)
  • Infrastructure is portable (fork the repo, switch to your own infra, keep developing)

2. Use Cases

2.1 Open Source Intercom for Your Service

intercom.png

The Use Case: Layer 0-1 support assistant embedded on your landing page or service, connected to your Telegram support group.

How It Works:

  • Webchatbot (AG-UI Protocol compatible) embedded on your site
  • Connects to AG-UI Gateway (SSE API) for real-time responses
  • Works as a Layer 0 AI agent with dynamic self-propelling knowledge base from FAQs and group responses
  • Escalates to Telegram support group with full context
  • Admins respond in Telegram; replies delivered back to user in web chat

Value:

support.png

  • Deflect support tickets with intelligent, contextual answers
  • Unified support across web and Telegram
  • No vendor lock-in (open source, self-hosted)

Example: DexGuru uses Atlas webchatbot to answer DeFi questions, route to support when needed, and surface on-chain data panels.

2.2 Chatbot with Comprehensive Workflows

The Use Case: Web/Telegram chat bot interface that guides users through complex scenarios (onboarding, quests, transactions) with forms, buttons, and conditional logic.

How It Works:

onboarding_flow.png

  • BPMN workflows define user journeys (visual diagrams or text/YAML) — Example BPMN | Example Forms
  • Forms engine (Camunda) renders interactive forms in bot and web
  • Sub-agents — Specialized agents with distinct roles, tools, and contexts work together in multi-agent systems
  • LLM personas (Context.md + YAML) customize agent behavior - Example Workflow with config.yaml
  • Task lists orchestrate user, LLM, and automated actions asynchronously

Value:

  • Predictable, testable user experiences (not just "prompt and hope")
  • Rich interactions (forms, buttons, inline keyboards, generated interfaces)
  • Business logic in sub agentic and BPMN workflows (not buried in code)

Example: Trip Planner bot creates trip artifacts through multi-step workflows; DexGuru scenarios drive users to token swaps with CTAs.

2.3 Gamification & Community Engagement

The Use Case: Quest systems, leaderboards, and achievement tracking that drive engagement and retention.

How It Works:

  • Quest workflows (BPMN) define missions with conditions and rewards
  • Leaderboards(example) track points, badges, and rankings
  • Cross-community competitions via Atlas Network
  • Wallet integration for token gating and rewards

Value:

  • Turn passive followers into active contributors
  • Measure and reward engagement
  • Network effects across communities

Example: Burning Meme (50K+ channel, 1K+ bot) uses high-frequency quests and hype loops to drive memecoin launches.

2.4 Community Knowledge Management

The Use Case: Self-propelling knowledge bases from Telegram groups and channels, with scheduled digests and on-demand answers.

How It Works:

  • Dynamic RAG — Streamed messages from groups/channels → Elasticsearch → searchable knowledge base
  • Scheduled LLM requests generate digests ("what you missed" recaps)
  • On-demand RAG queries answer questions with citations
  • Two types of RAG:
  • Group knowledge bases - Community conversations, FAQs, support threads
  • Dynamic inventory - Constantly updated data (tokens in DexGuru, points of interest in Trip Planner)

Value:

  • Knowledge bases that update themselves (no manual curation)
  • Context-aware answers with source attribution
  • Scheduled digests keep lurkers engaged

Example: Community bot ingests 1000+ messages/day, generates weekly digests, answers questions with citations from group history.

3. Main Use Cases (Out of the Box)

3.1 Knowledge Base Management with Streamed Data

What: Self-updating knowledge bases from Telegram groups/channels

How:

  • Messages → Elasticsearch → RAG queries
  • Scheduled digests ("what you missed" recaps)
  • On-demand answers with citations

Value:

  • No manual curation
  • Always fresh
  • Source attribution

3.2 Intercom-Style Assistant

What: Open-source support layer connecting web and Telegram

How:

  • Webchatbot embedded on site
  • Connects to AG-UI Gateway (AG-UI Protocol)
  • Escalates to Telegram support group
  • Admins respond; replies delivered back to user

Value:

  • Deflect support tickets
  • Unified support across channels
  • Self-hosted, no vendor lock-in

3.3 Tools Applied to Messages

What: Search, RAG, scheduled LLM, image processing, Whisper

How:

  • Messages trigger tool execution
  • Results fed back to LLM or user
  • Tools configurable via LangGraph or custom code

Value:

  • Rich interactions (not just text)
  • Automated processing (digests, summaries)
  • Extensible (add your own tools)

3.4 Comprehensive Dialog/Automated Workflows

What: Pre-set workflows with conditions, forks, forms, tools

How:

Value:

  • Predictable results (not "prompt and hope")
  • Business logic in workflows (not code)
  • Testable, version-controlled scenarios

4. Technical Stack

4.1 Core Framework

  • GURU Framework - Battle-tested infrastructure (search, evals, guardrails)
  • Camunda BPMN - Workflow orchestration engine (engine-api) - Modeler
  • LangGraph - AI layer (agentic workflows, tools, knowledge bases)
  • Elasticsearch - Dynamic RAG engine (StatefulSet)
  • Redis - Session and context store (StatefulSet)
  • PostgreSQL - Primary database (via PgBouncer, StatefulSet)
  • ClickHouse - Event bus and analytics (StatefulSet)
  • RabbitMQ - Async workloads and external tasks (StatefulSet)

4.2 Stateless Microservices (Kubernetes Deployments)

  • bot-app - Telegram bot + AG-UI Gateway (Python)
  • webchatbot-app - AI assistant for web (CopilotKit + AG-UI Protocol)
  • engine-api - BPMN workflow engine (Camunda 7)
  • flowapi-api - Auth, app config, analytics API
  • warehouse-api - WebSocket event stream
  • Workers - External Task Workers (RabbitMQ consumers)

4.3 Stateful Infrastructure (Kubernetes StatefulSets)

  • PostgreSQL (via PgBouncer) - Primary database for engine-api and flowapi-api
  • ClickHouse - Event bus and analytics storage
  • Redis - Cache and state machine (Sentinel or Cluster mode)
  • RabbitMQ - Async workloads (quorum queues)
  • Elasticsearch - Knowledge base search (multi-node cluster)

4.4 Infrastructure

  • Kubernetes - Container orchestration (namespaces per application)
  • Docker - Containerization
  • PgBouncer - PostgreSQL connection pooling
  • WebSocket - Real-time notifications (warehouse-api)
  • Service Mesh - Internal-only communication

4.5 Blockchain

  • Solana Web3.js - On-chain lookups
  • Wallet integration - Phantom, Solflare, Backpack
  • Token gating - Wallet-based access control

5. Examples & Resources

5.1 BPMN Workflows & Forms

Community Onboarding Example:

Tools:

  • Camunda Modeler — Visual BPMN editor for creating and editing workflows

5.2 Workflow Dialog Scenarios

Atlas Onboarding Workflow:

5.3 Open Source Projects

6. Development Workflow

6.1 Quick Start

  1. Launch via wizard - https://atlas.gurunetwork.ai/launcher
  2. Get env vars - /admin command in bot
  3. Fork repo - Switch to your own infrastructure
  4. Develop locally - ./run_development.sh
  5. Deploy - Push to your fork, update k8s configs

6.2 Customization

6.3 Best Practices

  • Start simple - Basic bot + RAG (camunda_enabled=false)
  • Add workflows - Enable Camunda for orchestration
  • Version control - Workflows, personas, configs in git
  • Test locally - Use run_development.sh against deployed infra
  • Iterate - Deploy, test, customize, repeat

7. Roadmap

7.1 Current (v0)

  • ✅ Dynamic RAG from streamed data
  • ✅ Intercom-style webchatbot
  • ✅ Basic workflows (Camunda optional)
  • ✅ Launcher wizard
  • ✅ AG-UI Protocol compatibility

7.2 Next (v1)

  • 🔄 Advanced workflows (multi-step, conditional)
  • 🔄 Custom tools marketplace
  • 🔄 Enhanced analytics (warehouse API)
  • 🔄 Cross-community features (Atlas Network)

7.3 Future (v2)

  • 📋 Multi-project workflows
  • 📋 Advanced agent personas (multi-agent systems)
  • 📋 Richer MCP/A2A tools
  • 📋 Enterprise features (SSO, audit logs, compliance)

Summary

Atlas is more than a chatbot framework - it's a complete orchestration platform for building AI-driven applications.

Key Innovations:

  • Dynamic RAG — Self-propelling knowledge bases from streamed data
  • Agentic Workflows - Comprehensive scenarios with forms, tools, and business logic
  • One-Click Deployment - Launcher wizard spins up complete applications in minutes
  • Portable Infrastructure - Fork repo, switch infra, keep developing

Use Cases: - Open-source Intercom for your service - Chatbot with comprehensive workflows - Gamification and community engagement - Community knowledge management

Get Started: - Launch your bot: https://atlas.gurunetwork.ai/launcher - Read the docs: https://atlas.gurunetwork.ai/docs - Fork the repo: https://github.com/evahteev/sol-atlas - Join the community: https://t.me/SolanaAtlas

Built on the GURU Framework • Powered by LangGraph • Connected by Solana