activepieces

Provides a React-based frontend UI that enables users to visually compose automation workflows by dragging action/trigger pieces onto a canvas and connecting them with data flow edges. The builder maintains a JSON-serialized flow definition that maps to the backend execution engine, with real-time validation of piece inputs/outputs and visual feedback for connection compatibility. State management via a centralized store tracks flow structure, piece configurations, and variable bindings.

Implements a plugin architecture where each integration (Discord, Google Drive, Claude, etc.) is a self-contained 'piece' package exporting actions and triggers via a standardized TypeScript interface. Pieces declare their inputs/outputs as JSON schemas, authentication requirements, and execution logic. The framework loads pieces dynamically at runtime via a piece-loader service that resolves dependencies, validates schemas, and injects authenticated connections from the connection management service.

Provides configurable error handling at the piece and flow level. Pieces can define error handlers that catch failures and trigger alternative actions. The execution engine supports automatic retries with exponential backoff (e.g., 1s, 2s, 4s, 8s) for transient failures. Retry logic is configurable per piece (max retries, backoff strategy). Failed steps can trigger error handlers that log, notify, or attempt recovery. Errors are tracked in the database for debugging and monitoring.

Provides a web-based UI for monitoring flow executions in real-time, showing step-by-step progress, intermediate outputs, and error details. The UI connects via WebSocket to the server's ProgressService, receiving live updates as steps execute. Users can inspect the output of each step, view variable values, and trace data flow through the workflow. Failed executions show detailed error messages and stack traces. The UI supports filtering and searching execution history.

Implements Activepieces as an MCP server, exposing flows and pieces as tools that AI agents (Claude, GPT, etc.) can invoke. Each piece is registered as an MCP tool with its JSON schema, allowing agents to discover available integrations and call them with natural language. The MCP server translates agent requests into flow executions, returning results back to the agent. This enables AI agents to autonomously execute multi-step workflows without explicit user orchestration.

Provides a translation system for the Activepieces UI, supporting multiple languages (English, Spanish, French, German, etc.). The frontend uses i18n libraries to load language-specific strings from JSON files and render the UI in the user's preferred language. Language selection is stored in user preferences and applied globally. The system supports right-to-left (RTL) languages and locale-specific formatting (dates, numbers, currency).

A TypeScript-based execution runtime (packages/engine) that interprets flow definitions as directed acyclic graphs, executing pieces sequentially or in parallel based on flow topology. The engine maintains execution context (FlowExecutionContext) tracking variables, step outputs, and execution state. It handles piece execution via PieceExecutor, code execution via CodeExecutor with sandboxing, loops via LoopExecutor, and conditional routing via RouterExecutor. Progress is tracked in real-time via a ProgressService and persisted to the database for resumability.

Exposes HTTP endpoints that accept incoming webhooks and map them to flow triggers. The webhook handler validates incoming payloads against the trigger's JSON schema, extracts relevant data, and enqueues a flow execution job with the webhook payload as the trigger input. Supports multiple webhook URLs per flow for different trigger types. Webhooks are authenticated via API keys or OAuth tokens depending on the flow's security configuration.

Implements a job queue (using Bull or similar) where flow executions are enqueued as jobs and processed by worker processes. The server enqueues execution jobs; workers dequeue and execute them using the flow execution engine. This decouples flow submission from execution, enabling horizontal scaling by adding more workers. Workers report progress back to the server via the ProgressService, which updates the database and broadcasts to connected clients via WebSockets.

Provides a centralized connection management system where users authenticate once to a service (via OAuth, API key, or basic auth) and the credentials are stored encrypted in the database. Pieces reference connections by ID, and the ConnectionService injects authenticated clients into piece execution. Supports multiple authentication methods per service (e.g., OAuth for user-scoped access, API key for service accounts). Credentials are encrypted at rest using the platform's encryption key.

Allows users to write custom JavaScript code blocks within workflows that execute in an isolated sandbox (using vm2 or similar). The sandbox has access to the flow execution context (variables, step outputs) but is restricted from accessing the file system, network, or other sensitive resources. Code blocks are executed with a 5-second timeout and can return values that flow into subsequent steps. Errors in code blocks are caught and can trigger error handlers.

Supports multiple trigger types for activating flows: webhook triggers (HTTP POST), polling triggers (periodic checks), and manual triggers (user-initiated). Polling triggers are executed by a scheduler that runs at configured intervals (e.g., every 5 minutes) and checks for new data. Triggers are implemented as pieces that export a trigger interface with a polling function or webhook handler. The flow engine enqueues a new execution for each trigger event.

Maintains version history for flows, allowing users to view, compare, and revert to previous versions. Flows are stored as JSON definitions in the database with version metadata (timestamp, author, change description). The platform supports git sync, enabling flows to be exported to a git repository and imported back, facilitating version control and CI/CD integration. Deployments can be promoted from development to production environments with approval workflows.

Implements a quota system that tracks usage metrics (flow executions, data processed, API calls) per user or workspace and enforces limits based on the subscription tier. The billing service records usage events, calculates charges, and prevents execution if quotas are exceeded. Quotas are configurable per plan (free, pro, enterprise) and can be soft limits (warnings) or hard limits (blocking execution). Usage is tracked in real-time via the execution engine and aggregated for billing.