> ## Documentation Index > Fetch the complete documentation index at: https://www.siya.com/llms.txt > Use this file to discover all available pages before exploring further. # Agent System Architecture > Learn about Siya's adaptive multi-agent system and specialized capabilities Siya employs a sophisticated multi-agent architecture that adapts based on your chosen execution mode. At the core is the main Siya agent, which either directly handles tasks using tools and sub-agents (Chat Mode) or orchestrates specialized agent modules (Task Mode). ## Agent System Overview ```mermaid theme={null} graph TB subgraph "SIYA AGENT ECOSYSTEM" SA["🤖 Main SIYA Agent
(Orchestrator)"] subgraph "Specialized Agents" SWE["🛠️ SWE Agent
Software Engineering"] Search["🔍 Search Agent
Research & Analysis"] Browser["🌐 Browser Agent
Web Automation"] Auto["🔄 Automation Agent
Task Scheduling"] end subgraph "Core Capabilities" Tools["🔧 Direct Tools"] Memory["💾 Memory System"] MCP["🔌 MCP Servers"] Hooks["🪝 Hooks System"] end SA --> SWE SA --> Search SA --> Browser SA --> Auto SA --> Tools SA --> Memory SWE --> MCP Tools --> Hooks end style SA fill:#0000FF,stroke:#fff,color:#fff style SWE fill:#07C983,stroke:#333,stroke-width:2px style Search fill:#15803D,stroke:#333,stroke-width:2px style Browser fill:#FF9800,stroke:#333,stroke-width:2px style Auto fill:#9C27B0,stroke:#333,stroke-width:2px ``` ## Core Agent Types The central intelligence that coordinates all operations and makes high-level decisions Understands user intent and determines the best approach Decides which tools or sub-agents to employ Manages communication between different components Ensures outputs meet requirements and standards ```mermaid theme={null} flowchart LR Input["User Input"] --> Analysis["Intent Analysis"] Analysis --> Complexity{Complexity?} Complexity -->|Simple| Direct["Direct Tool Use"] Complexity -->|Complex| Delegate["Delegate to Sub-Agent"] Complexity -->|Multi-Domain| Multiple["Multiple Agents"] Direct --> Execute["Execute & Return"] Delegate --> Monitor["Monitor Progress"] Multiple --> Coordinate["Coordinate Agents"] Execute --> Result["Final Result"] Monitor --> Result Coordinate --> Result ```
Capability Chat Mode Task Mode
Direct Tool Access ✅ Full Access ❌ Delegates Only
Sub-Agent Spawning ✅ On-Demand ✅ Pre-Planned
User Interaction ✅ Continuous ⚡ Minimal
Planning Depth 🔄 Adaptive 📊 Comprehensive
Your dedicated software development partner with deep technical expertise * Architecture design * Code generation * Testing & debugging * Performance optimization * Security analysis * Documentation * MCP server access * Build system integration * Multi-language support * Git operations * CI/CD pipeline setup * Code review capabilities
**Frontend** * JavaScript/TypeScript * React/Vue/Angular * HTML/CSS * WebAssembly
**Backend** * Python * Node.js * Java/Kotlin * Go * Rust * C/C++
**Specialized** * SQL/NoSQL * GraphQL * Shell scripting * Configuration files * Docker/K8s
Comprehensive knowledge of modern frameworks: * **Web**: Express, Django, Spring Boot, FastAPI * **Mobile**: React Native, Flutter, SwiftUI * **Data**: Pandas, NumPy, TensorFlow, PyTorch * **Testing**: Jest, Pytest, JUnit, Selenium * **DevOps**: Terraform, Ansible, GitHub Actions VSCode, IntelliJ, Vim configurations Git, GitHub, GitLab, Bitbucket Webpack, Gradle, Maven, Make Docker, Kubernetes, Helm The SWE Agent has exclusive access to MCP servers in Task Mode, enabling powerful integrations * PostgreSQL operations * MongoDB queries * Redis caching * Elasticsearch indexing * API integrations * Cloud services * Third-party tools * Custom protocols ```mermaid theme={null} sequenceDiagram participant U as User participant M as Main Agent participant S as SWE Agent participant MCP as MCP Server participant G as Git U->>M: "Create a REST API" M->>S: Delegate task S->>S: Design architecture S->>S: Generate code S->>MCP: Set up database MCP-->>S: DB configured S->>S: Write tests S->>G: Commit changes S-->>M: Task complete M-->>U: API ready with docs ``` Advanced research capabilities with multi-source synthesis and fact verification Searches across web, databases, and documentation simultaneously Combines findings from multiple sources into coherent insights Cross-references information for accuracy and reliability Provides proper sources and references for all findings
**Technical Research** * API documentation * Library comparisons * Best practices * Performance benchmarks * Security advisories * Version compatibility
**General Research** * Market analysis * Competitive intelligence * Trend identification * Statistical data * Academic papers * Industry reports
Explores multiple sources quickly for comprehensive overview **Best for:** * Initial exploration * Trend analysis * Comparative studies * Market research Focuses intensely on specific sources for detailed information **Best for:** * Technical specifications * Academic research * Root cause analysis * Expert knowledge Progressively narrows focus based on initial findings **Best for:** * Complex topics * Evolving requirements * Precision needs * Quality over quantity * Executive summaries * Detailed analyses * Comparison tables * Recommendation lists * Structured datasets * Statistical summaries * Trend graphs * Raw findings * Annotated bibliographies * Source rankings * Citation formats * Reading lists Visual AI that interacts with websites like a human, handling complex web workflows The Browser Agent uses computer vision to interact with web pages, making it capable of handling dynamic content and visual challenges that traditional automation tools cannot.
**Visual Understanding** * Page layout analysis * Button/link recognition * Form field detection * Dynamic content tracking * Image/text extraction * Color/style awareness
**Interaction Skills** * Click precision * Text input * Dropdown navigation * File uploads * Drag and drop * Scroll control
Handles complex sequences like checkout processes, form wizards, and multi-page operations Adjusts to different site layouts and unexpected changes Maintains cookies, authentication, and state across interactions Handles popups, loading delays, and unexpected interface changes ```mermaid theme={null} graph LR A[Search Product] --> B[Add to Cart] B --> C[Checkout] C --> D[Fill Shipping] D --> E[Payment Info] E --> F[Confirm Order] F --> G[Get Confirmation] style A fill:#FF9800,stroke:#333 style G fill:#4CAF50,stroke:#333 ``` * Product searches * Price comparisons * Inventory checks * Order placement * Tracking updates * Table data extraction * Pagination handling * Dynamic content capture * Image downloading * PDF generation Job applications, registrations, surveys Bulk form filling from spreadsheets CAPTCHA handling, email confirmation Browser Agent Architecture The Browser Agent maintains its own browser context, separate from your system browser, ensuring privacy and isolation. Creates and manages automated workflows, scheduled tasks, and system integrations * Cron-like scheduling * One-time delays * Recurring patterns * Business hours aware * Timezone handling * Holiday calendars * Multi-step processes * Conditional logic * Error handling * Parallel execution * State management * Result aggregation ```text Recurring theme={null} "Every day at 9 AM" "Weekly on Mondays" "First Friday of month" "Every 2 hours" "Weekdays at 5 PM" ``` ```text One-Time theme={null} "Tomorrow at 3 PM" "In 30 minutes" "Next Monday" "December 25 at noon" "After market close" ``` ```text Complex theme={null} "Every MWF at 10 AM" "Hourly between 9-5" "Last day of quarter" "When file changes" "On system startup" ``` Gather data from multiple sources Clean, format, and enrich data Generate insights and metrics Send reports to stakeholders * Log rotation * Backup creation * Cache clearing * Health checks * Resource monitoring * Alert generation
**Source Systems** * Databases * APIs * File systems * Email * Webhooks
**Target Actions** * Notifications * Data updates * Report generation * Task creation * System commands
```mermaid theme={null} flowchart TD Trigger["⏰ Trigger
(Schedule/Event)"] --> Check{Conditions
Met?} Check -->|Yes| Action1["📊 Fetch Data"] Check -->|No| Wait["⏳ Wait/Skip"] Action1 --> Process["⚙️ Process"] Process --> Validate{Valid?} Validate -->|Yes| Notify["📧 Notify"] Validate -->|No| Error["❌ Error Handler"] Notify --> Log["📝 Log Results"] Error --> Log Log --> Next["🔄 Next Execution"] style Trigger fill:#9C27B0,stroke:#fff,color:#fff style Notify fill:#4CAF50,stroke:#333 ``` * Deploy on commit * Test suite runs * Performance monitoring * Incident response * Sales reports * Invoice processing * Customer follow-ups * Inventory alerts * Email summaries * Task reminders * Data backups * Health tracking ## Agent Communication Patterns Agents communicate through well-defined protocols, ensuring efficient collaboration and data sharing ```mermaid theme={null} sequenceDiagram participant User participant Main as Main SIYA Agent participant SWE as SWE Agent participant Search as Search Agent participant Browser as Browser Agent participant Memory as Shared Memory User->>Main: Complex request Main->>Main: Analyze & Plan par Parallel Execution Main->>Search: Research requirements and Main->>SWE: Prepare environment end Search->>Memory: Store findings SWE->>Memory: Update progress Main->>Browser: Execute web tasks Browser->>Memory: Save results Main->>Memory: Aggregate all data Memory-->>Main: Combined results Main-->>User: Final deliverable Note over Main,Memory: Continuous synchronization ``` ## Agent Selection Logic Siya automatically selects the most appropriate agent(s) based on task analysis
Task Type Primary Agent Supporting Agents
Build a web app SWE Agent Search (for docs)
Market research Search Agent Browser (for data)
Form submission Browser Agent -
Daily reports Automation Agent Search, SWE
Debug code SWE Agent Search (for errors)
You can explicitly request specific agents in Chat Mode ```text SWE Request theme={null} "Use the SWE agent to refactor this code" ``` ```text Search Request theme={null} "Have the search agent find the latest React best practices" ``` ```text Browser Request theme={null} "Use the browser agent to fill out this form" ``` ```text Multiple Agents theme={null} "Use search agent for research and SWE agent for implementation" ``` ## Performance & Optimization Agents can work simultaneously on independent tasks, dramatically reducing total execution time Intelligent resource allocation ensures optimal performance without system overload
Operation Single Agent Multi-Agent Improvement
Full-stack app 45 min 15 min 3x faster
Research report 30 min 8 min 3.75x faster
Data pipeline 60 min 20 min 3x faster
**CPU Usage** * Main Agent: 5-10% * SWE Agent: 10-20% * Search Agent: 5-15% * Browser Agent: 15-25% * Automation: 5-10%
**Memory Usage** * Base system: 500MB * Per agent: 200-400MB * Shared memory: 1GB * Peak usage: 3-4GB
Pre-planned execution is more efficient than interactive guidance Group related tasks for better resource utilization Reset conversation when switching to unrelated tasks Use lighter models for simple tasks, powerful ones for complex work ## Agent Capabilities Matrix Understanding what each agent can do helps you leverage Siya's full potential
Capability Main SWE Search Browser Auto
File Operations ✅ ✅ ❌ ⬇️ ✅
Code Generation ✅ ✅ ❌ ❌ ❌
Web Research ✅ 📚 ✅ ✅ ❌
API Calls ✅ ✅ ✅ ❌ ✅
Visual Interaction ❌ ❌ ❌ ✅ ❌
Scheduling 📅 ❌ ❌ ❌ ✅
MCP Access ❌ ✅ ❌ ❌ ❌
✅ Full capability | 📚 Limited capability | ⬇️ Download only | 📅 Basic only | ❌ Not available ## Best Practices * Let Siya choose automatically when unsure * Use specific agents for specialized tasks * Combine agents for complex projects * Consider mode implications * Batch similar operations together * Use parallel execution when possible * Clear memory between major tasks * Monitor resource usage ## Advanced Agent Features Agent state and context can be preserved across sessions in certain scenarios * **Chat Mode**: Conversation history maintained * **Task Mode**: Execution logs preserved * **Automation**: Schedules persist indefinitely * **Workspace**: Files and data remain available ```json theme={null} { "agents": { "swe": { "timeout": 300, "maxRetries": 3, "preferredLanguages": ["python", "typescript"], "gitAutoCommit": false }, "search": { "maxResults": 20, "searchDepth": 3, "verifyFacts": true }, "browser": { "headless": true, "viewport": "1920x1080", "waitTimeout": 30 } } } ``` * Custom agent creation * Third-party agent integration * Agent marketplace * Specialized domain agents ## Real-World Examples ```mermaid theme={null} gantt title Full-Stack App Development Timeline dateFormat HH:mm section Research Market Research :done, search1, 00:00, 15m Tech Stack Research :done, search2, 00:05, 10m section Development Backend API :active, swe1, 00:15, 30m Frontend UI :active, swe2, 00:20, 25m Database Setup :active, swe3, 00:15, 20m section Testing Unit Tests :browser1, 00:45, 15m Integration Tests :browser2, 00:50, 20m section Deployment CI/CD Setup :auto1, 01:00, 15m Monitoring :auto2, 01:10, 10m ``` **Search Agent** gathers data sources and **Browser Agent** extracts from websites **SWE Agent** writes processing scripts and handles transformations **SWE Agent** performs statistical analysis and generates visualizations **Automation Agent** schedules daily runs and report distribution A comprehensive research project leveraging all agents
**Phase 1: Discovery** * Search Agent explores topics * Browser Agent accesses papers * 20+ sources analyzed
**Phase 2: Synthesis** * SWE Agent processes data * Creates visualizations * Generates citations
**Phase 3: Validation** * Search Agent fact-checks * Browser Agent verifies sources * Cross-references findings
**Phase 4: Delivery** * SWE Agent formats report * Automation Agent schedules updates * Delivers to stakeholders
## Troubleshooting
Issue Cause Solution
Agent timeout Complex task Break into smaller subtasks
Memory overflow Large context Clear conversation history
Slow performance Multiple agents Use Task Mode for efficiency
Wrong agent selected Ambiguous request Specify agent explicitly
View current agent activity in the status bar Access detailed execution logs for debugging "Why did you choose that agent?" gets explanations Email [dev@siya.com](mailto:dev@siya.com) for persistent issues ## Summary Siya's agent system provides specialized expertise for every task. Understanding each agent's strengths and how they work together unlocks incredible productivity gains. Start with automatic selection and gradually learn to orchestrate agents for maximum efficiency. Explore the specific tools available to agents Open Siya and experiment with different agents *** *Specialized agents. Coordinated intelligence. Unlimited possibilities.*