Core Concepts

Beam is built on key concepts that work together to create powerful AI automation. Understanding these fundamentals will help you build more effective agents and workflows.

​

Agents

Agents are autonomous AI systems that execute tasks on your behalf. Each agent operates independently with:

• Purpose: Specific job or workflow designed to handle
• Intelligence: LLM-powered decision making and reasoning
• Tools: Access to integrations and capabilities
• Memory: Context awareness across tasks and conversations
• Personality: Configurable behavior and response style

Agents can handle complex workflows, learn from interactions, and adapt their behavior based on context and feedback.

​

Flows (Graphs)

Flows define the step-by-step logic your agent follows to complete tasks. They are visual workflows represented as graphs with:

​

Graph Components

• Nodes: Individual workflow steps with specific objectives
• Edges: Connections defining execution flow between nodes
• Conditions: Branching logic for dynamic workflows
• Tools: Actions performed at each node

​

Node Types

Flows support linear sequences, conditional branches, parallel execution, and loops for complex automation patterns.

​

Tools & Integrations

Tools give your agents capabilities to interact with external systems and perform actions:

​

Tool Categories

System Tools: Built-in platform capabilities for database operations, file management, and communication Integration Tools: Pre-built connectors for 1500+ services including Gmail, Salesforce, Slack, Google Workspace, and Microsoft 365 Custom Tools: User-created business logic and proprietary algorithms tailored to specific needs AI-Generated Tools: Tools created through natural language descriptions with automatic parameter detection Tools handle authentication, rate limiting, error handling, and retry logic automatically.

​

Tasks

Tasks are execution instances of your agent. When you run an agent:

​

Task Lifecycle

​

Task States

• PENDING: Awaiting execution
• RUNNING: Currently processing
• PAUSED: Waiting for user input or consent
• COMPLETED: Successfully finished
• FAILED: Encountered error
• CANCELLED: User terminated

​

Variables & State

Variables maintain context throughout your agent’s execution:

• Input Variables: Data provided when starting a task
• Intermediate Variables: Values computed during execution
• Output Variables: Results returned to the user
• Persistent State: Memory carried across multiple tasks

Variables can be strings, numbers, objects, arrays, or structured data matching custom schemas.

​

Memory System

Memory enables agents to remember and learn from interactions:

​

Memory Types

Short-term Memory: Current task context and active conversation Long-term Memory: Persistent knowledge accumulated over time Working Memory: Active processing and temporary data Episodic Memory: Sequences of events and past interactions Memory uses vector embeddings for semantic search and intelligent retrieval of relevant context.

​

Triggers

Triggers define how agents start execution: Triggers can include filters and conditions to process only relevant events.

​

Workspaces

Workspaces organize your team’s automation:

• Shared Resources: Agents, integrations, and tools accessible to team members
• Access Control: Role-based permissions (Admin, Editor, Viewer)
• Centralized Management: Billing, usage tracking, and settings
• Multi-tenant Isolation: Secure separation between workspaces

Workspaces enable collaboration while maintaining security and resource boundaries.

​

Authentication & Integrations

Beam handles authentication for all integrations:

​

Authentication Methods

• OAuth 2.0: Secure authorization for Google, Microsoft, Slack, and more
• API Keys: Direct API access with key management
• Custom Auth: Support for proprietary authentication schemes

Integration credentials are securely stored, automatically refreshed, and never exposed to agents.

​

Platform Architecture

Beam consists of three interconnected components:

​

Next Steps