n8n

Provides a drag-and-drop canvas interface for building directed acyclic graphs (DAGs) of interconnected nodes representing integrations and data transformations. The frontend uses Vue.js state management to track node positions, connections, and parameter configurations in real-time, with expression evaluation for dynamic values. Users can visually wire outputs from one node to inputs of another, with the system automatically managing data flow and type inference across the graph.

Implements a sandboxed JavaScript expression evaluator (via @n8n/expression-runtime package) that allows users to write inline expressions in node parameters using a custom syntax with access to workflow context, previous node outputs, and utility functions. Expressions are parsed, validated, and executed within an isolated runtime that prevents arbitrary code execution while providing access to $node, $json, $env, and other context variables. This enables dynamic parameter values without requiring separate code nodes.

Maintains a persistent execution history database storing details of every workflow execution including start/end times, status, node-level logs, input/output data, and error messages. Executions are indexed and searchable by workflow ID, status, date range, and error type. The system provides APIs and UI views for inspecting execution history, filtering by criteria, and exporting logs for compliance or debugging. Supports configurable retention policies to manage database size.

Implements a project-based authorization model where workflows are organized into projects with granular permission controls (view, edit, execute, admin). Users can be assigned roles at the project level, and workflows inherit permissions from their parent project. The system supports team collaboration with shared projects, audit logging of permission changes, and optional SSO integration for enterprise deployments. Credentials are scoped to projects and can be shared across workflows within the same project.

Provides error handling mechanisms including try-catch nodes, error output branches, and configurable retry policies with exponential backoff. When a node fails, the workflow can route to an error handler node, retry the failed node with increasing delays, or halt execution. Retry policies are configurable per node with parameters for max attempts, initial delay, backoff multiplier, and maximum delay. Failed executions are logged with error details and can trigger notifications or escalations.

Provides a Data Store node that allows workflows to persist and retrieve key-value data across multiple executions. Data is stored in the n8n database and can be accessed by any workflow with appropriate permissions. Supports operations like set, get, delete, and list with optional TTL (time-to-live) for automatic expiration. The data store enables workflows to maintain state between executions, implement counters, or cache frequently accessed data.

Supports workflow versioning where each workflow modification creates a new version with automatic or manual snapshots. Workflows can be exported to JSON and imported from version control systems (Git), enabling CI/CD integration. The system tracks version history with timestamps and user information, allowing rollback to previous versions. Integration with Git repositories enables collaborative development with branch-based workflows and pull request reviews.

Implements multi-tenancy where each organization/workspace has isolated workflows, credentials, and execution history. Credentials are encrypted and scoped to specific tenants, preventing cross-tenant access. Execution isolation ensures that workflows from different tenants don't interfere with each other, with separate execution queues and resource limits per tenant. The system supports tenant-level configuration including custom branding, feature flags, and API rate limits.

Maintains a registry of 400+ integration nodes (HTTP, Slack, Salesforce, Stripe, etc.) defined in the @n8n/nodes-base package, each with a standardized node interface that includes input/output definitions, parameter schemas, and credential bindings. The credential system uses encrypted storage with support for OAuth2, API keys, and custom authentication schemes. Nodes are dynamically loaded at runtime and registered in a NodeTypeManager, allowing users to select integrations from a searchable catalog and configure them with stored or new credentials.

Implements a distributed execution model where the main n8n process can delegate workflow execution to separate task-runner processes via a message queue (Redis or Bull). The WorkflowExecutor in @n8n/core orchestrates execution across multiple runners, managing job queuing, result aggregation, and failure recovery. Supports both single-process execution (for development) and distributed execution (for production), with automatic load balancing and horizontal scaling by adding more task-runner instances.

Integrates LangChain nodes and an AI Workflow Builder that accepts natural language descriptions and generates n8n workflow definitions using LLM prompting. The system parses user intent, maps it to available nodes and integrations, and constructs a valid DAG structure with parameter bindings. Supports iterative refinement where users can ask the AI to modify generated workflows, with the system maintaining context across conversation turns.

Provides an Instance AI node that runs an autonomous LLM agent within a workflow, allowing the agent to use available n8n nodes as tools to accomplish goals. The agent receives a task description, iteratively selects and executes nodes based on their descriptions and outputs, and continues until the goal is achieved or a maximum iteration limit is reached. The system maintains agent state across iterations and provides execution logs showing the agent's reasoning and tool selections.

Provides a Webhook node that exposes HTTP endpoints for triggering workflow execution. Each webhook generates a unique URL that accepts POST/GET requests with arbitrary JSON payloads. The system validates incoming payloads against optional JSON schema definitions, filters requests based on headers or body conditions, and queues workflow executions with the payload as initial context. Supports both synchronous (wait for result) and asynchronous (fire-and-forget) execution modes.

Implements a Trigger node that executes workflows on a schedule using cron expressions (e.g., '0 9 * * MON-FRI' for 9 AM weekdays). The scheduler runs in the main n8n process or delegates to task runners, supporting timezone-aware scheduling with automatic DST handling. Users can configure execution frequency, timezone, and optional payload data to pass to the workflow on each execution.

Provides a Code node that allows users to write arbitrary Node.js JavaScript code that executes within the workflow. The code has access to the workflow context ($node, $json, $env), can import npm packages (if configured), and can perform complex data transformations or logic that's not available in pre-built nodes. Code is executed in a sandboxed V8 context with configurable timeout and memory limits to prevent resource exhaustion.

Implements conditional execution flow using IF and SWITCH nodes that evaluate expressions and route execution to different branches based on results. The IF node supports simple true/false conditions, while SWITCH node supports multiple conditions with fallback routing. Conditions are evaluated using the expression engine, allowing complex logic based on previous node outputs, variables, and environment data. Execution follows only the matching branch, skipping other branches.