Architecture
How the isA platform components work together.
For the enterprise consumption and governance layer above these providers, see the AEP control plane guide. AEP coordinates projects, capability requests, policy, approvals, evidence, and release intent; the provider remains authoritative for runtime execution and source-system truth.
Platform Overview
┌─────────────────────────────────────────────────────────────────────────────────┐
│ APPLICATIONS │
│ Web Apps • Mobile Apps • CLI Tools • Integrations │
└─────────────────────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────────┐
│ AGENT SDK (Python) │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐│
│ │ Agent │ │ Workflow │ │ Memory │ │ Tools │ │ Voice │ │ A2A ││
│ │ Runtime │ │ Engine │ │ Manager │ │ Registry │ │ Service │ │Protocol││
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘ └────────┘│
└─────────────────────────────────────────────────────────────────────────────────┘
│
┌───────────────────┼───────────────────┐
▼ ▼ ▼
┌──────────────────────┐ ┌──────────────────────┐ ┌──────────────────────┐
│ MCP SERVICE │ │ MODEL SERVICE │ │ USER SERVICES │
│ │ │ │ │ │
│ 190+ Tools: │ │ LLM Gateway: │ │ 35 Microservices: │
│ • Web Search │ │ • 8+ Providers │ │ • Authentication │
│ • File Operations │ │ • Role Routing │ │ • Payments │
│ • Code Execution │ │ • Voice/Realtime │ │ • Storage │
│ • Marketplace │ │ • ML Prediction │ │ • Organizations │
│ • Browser │ │ • Caching │ │ • Devices (IoT) │
│ • APIs │ │ • Cost Ranking │ │ • Memory │
└──────────────────────┘ └──────────────────────┘ └──────────────────────┘
│ │ │
└───────────────────┼───────────────────┘
│
┌───────────────────┼───────────────────┐
▼ ▼
┌──────────────────────────────────┐ ┌──────────────────────────────────┐
│ DOMAIN SERVICES │ │ PLATFORM TOOLING │
│ │ │ │
│ ┌──────────────┐ ┌───────────┐ │ │ ┌──────────────────────┐ │
│ │ Creative │ │ Marketing │ │ │ │ Orchestration │ │
│ │ │ │ │ │ │ │ (isA Orch) │ │
│ │ • Content │ │ • Campaign│ │ │ │ │ │
│ │ Creator │ │ Manager │ │ │ │ • Unified Code Index │ │
│ │ • Visual │ │ • SEO │ │ │ │ • Project Registry │ │
│ │ Designer │ │ Agent │ │ │ │ • Impact Analysis │ │
│ │ • Brand │ │ • Content │ │ │ │ • Cross-Project │ │
│ │ Strategist │ │ Agent │ │ │ │ Search │ │
│ │ • Copy Editor│ │ • Audience│ │ │ └──────────────────────┘ │
│ │ • DAM Manager│ │ Insights│ │ │ │
│ │ • Design │ │ • Campaign│ │ │ │
│ │ System │ │ Analyst │ │ │ │
│ └──────────────┘ └───────────┘ │ │ │
└──────────────────────────────────┘ └──────────────────────────────────┘
│ │
└───────────────────┬───────────────────┘
▼
┌─────────────────────────────────────────────────────────────────────────────────┐
│ DATA SERVICE │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Data Lake │ │ Data Fabric │ │ RAG │ │ File │ │
│ │ (DuckDB) │ │ (Pipelines) │ │ (Vectors) │ │ Processing │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘ │
└─────────────────────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────────┐
│ isA CLOUD │
│ │
│ ┌────────────────────────────────────────────────────────────────────────┐ │
│ │ APISIX GATEWAY (Port 9080) │ │
│ │ Rate Limiting • Auth • Routing • SSL • Metrics │ │
│ └────────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ┌────────────────────────────────────────────────────────────────────────┐ │
│ │ CONSUL SERVICE DISCOVERY (Port 8500) │ │
│ │ Registration • Health Checks • DNS • KV Store │ │
│ └────────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ┌────────────────────────────────────────────────────────────────────────┐ │
│ │ gRPC SERVICES (50051-50070) │ │
│ │ postgres • redis • neo4j • nats • mqtt • minio • qdrant • duckdb │ │
│ └────────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ┌────────────────────────────────────────────────────────────────────────┐ │
│ │ INFRASTRUCTURE │ │
│ │ PostgreSQL • Redis • Neo4j • NATS • MQTT • MinIO • Qdrant • DuckDB │ │
│ └────────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────────┘Component Details
Agent SDK
The Python SDK is the primary interface for building agents:
| Component | Purpose |
|---|---|
| Agent Runtime | Executes agent logic, manages state |
| Workflow Engine | Swarm handoffs and DAG orchestration |
| Memory Manager | Handles short and long-term memory |
| Tools Registry | Discovers and invokes tools |
| Voice Service | Full-duplex voice with audio buffering and STT rewriting |
| A2A Protocol | Agent-to-agent communication via Agent Card + JSON-RPC |
| Delegation | Remote team task delegation with circuit breakers |
| Observability | Prometheus metrics for nodes, models, tools, and pools |
MCP Service
Model Context Protocol provides 190+ tools with marketplace and multi-tenant security:
| Category | Tools |
|---|---|
| Search | web_search, image_search, news_search |
| Files | file_read, file_write, file_list |
| Code | code_interpreter, shell, git |
| Data | database_query, api_call, scrape |
| Browser | browser_navigate, browser_click |
| Marketplace | Package install, update, enable/disable |
| Security | JWT/API key auth, command allowlist, SSE injection prevention |
Model Service
Unified LLM gateway with intelligent routing:
| Feature | Description |
|---|---|
| Multi-Provider | OpenAI, Anthropic, OpenRouter, Cerebras, Ollama, YYDS, Replicate, Anthropic Sub |
| Intelligent Routing | Role-based resolution, capability filtering, cost-aware ranking |
| Voice/Realtime | WebSocket sessions with privacy filter and turn detection |
| ML Prediction | Time-series forecasting with Prophet (training + inference) |
| Caching | L1 exact-match (Redis) + L2 semantic (Qdrant) |
| Failover | Automatic provider failover with circuit breakers |
User Services
35 microservices for common application needs:
| Category | Services |
|---|---|
| Auth | auth, profile, rbac, social |
| Payments | payment, subscription, billing, wallet |
| Storage | storage, album, media, document |
| Social | organization, invitation, membership |
| IoT | device, ota, telemetry |
| AI | memory (cognitive memory system) |
Creative Service
AI-powered creative agents for content production and brand management:
| Agent | Purpose |
|---|---|
| Content Creator | Generates written content (articles, posts, scripts) |
| Visual Designer | Produces visual assets and design concepts |
| Brand Strategist | Maintains brand voice and positioning |
| Copy Editor | Reviews and refines content for quality |
| DAM Manager | Digital asset management and organization |
| Design System Agent | Enforces design system consistency |
Marketing Service
AI-driven marketing automation and analytics:
| Agent | Purpose |
|---|---|
| Campaign Manager | Plans and executes marketing campaigns |
| SEO Agent | Search engine optimization and keyword strategy |
| Content Agent | Marketing-specific content generation |
| Audience Insights | Audience segmentation and analysis |
| Campaign Analyst | Campaign performance analytics and reporting |
| Compliance Agent | Ensures marketing content meets regulations |
| Report Generator | Automated marketing reports and dashboards |
Orchestration (isA Orch)
Platform-level tooling for cross-project operations:
| Component | Purpose |
|---|---|
| Unified Code Index | Single index spanning all isA projects with symbol resolution |
| Project Registry | Configuration and relationships of all sub-projects |
| Platform CLI | Cross-project indexing, search, and impact analysis |
Data Service
Data processing and analytics:
| Component | Technology |
|---|---|
| Data Lake | DuckDB for analytics |
| Data Fabric | Apache Spark pipelines |
| RAG | Qdrant + pgvector |
| File Processing | Multi-format extraction |
isA Cloud
Production infrastructure:
| Component | Technology | Port |
|---|---|---|
| Gateway | Apache APISIX | 9080 |
| Discovery | Consul | 8500 |
| Database | PostgreSQL | 5432 |
| Cache | Redis | 6379 |
| Graph | Neo4j | 7687 |
| Messaging | NATS | 4222 |
| IoT | MQTT | 1883 |
| Storage | MinIO | 9000 |
| Vectors | Qdrant | 6333 |
| Analytics | DuckDB | - |
Data Flow
Agent Request Flow
1. User sends request to Agent
2. Agent SDK processes request
3. Agent decides which tools to use
4. Tools call backend services via MCP
5. Model service generates responses
6. Results aggregated and returnedAuthentication Flow
1. User authenticates via auth_service
2. JWT token issued
3. Token included in API requests
4. APISIX validates token
5. Request routed to service
6. Response returnedEvent Flow
1. Service publishes event to NATS
2. Interested services subscribe
3. Events processed asynchronously
4. State updated across services
5. Notifications sent if neededDeployment Options
Local Development
# Using KIND (Kubernetes in Docker)
cd deployments/kubernetes/local/scripts
./kind-setup.sh
./kind-deploy.shStaging/Production
# GitOps with ArgoCD
git push origin main # Triggers deployment| Environment | Cluster | Auto-Deploy |
|---|---|---|
| Local | KIND | Manual |
| Staging | EKS | Yes (main branch) |
| Production | EKS/GKE | Manual approval |
Security
Authentication
- JWT tokens for API access
- OAuth2/OIDC for SSO
- API keys for service-to-service
Authorization
- RBAC with fine-grained permissions
- Organization-level isolation
- Resource-level access control
Data Protection
- Encryption at rest (AES-256)
- Encryption in transit (TLS 1.3)
- Secrets in Vault/AWS Secrets Manager
Next Steps
- Getting Started - Build your first agent
- Cloud Setup - Deploy infrastructure
- Security Guide - Authentication details