n8n-io/n8n

Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.

Structural Verdict

How Data Flows Through the System

Workflows are created in the frontend editor, stored as JSON definitions, executed by the core engine, and can trigger external integrations or AI agents

1. Workflow Creation — User builds workflow in Vue editor or via API (config: N8N_PERSONALIZATION_ENABLED, N8N_AI_ENABLED)
2. Workflow Storage — JSON workflow definition stored in database
3. Execution Trigger — Manual, webhook, or scheduled trigger starts execution (config: EXTERNAL_FRONTEND_HOOKS_URLS)
4. Node Execution — Core engine executes nodes sequentially based on connections (config: task_timeout, max_concurrency)
5. AI Processing — AI nodes process using configured models and memory (config: stdlib_allow, external_allow)
6. Result Output — Execution results stored and returned to client (config: SKIP_STATISTICS_EVENTS)

System Behavior

How the system actually operates at runtime — where data accumulates, what loops, what waits, and what controls what.

Data Pools

Feedback Loops

• Workflow Retry Logic (retry, balancing) — Trigger: Node execution failure. Action: Retry node with exponential backoff. Exit: Max retries reached or success.
• AI Agent Iteration (recursive, reinforcing) — Trigger: Agent needs additional tool calls. Action: Execute tool and feed result back to model. Exit: Final answer or max iterations.
• Task Runner Health Check (polling, balancing) — Trigger: Health check interval. Action: Ping Python task runner service. Exit: Service shutdown.

Delays & Async Processing

• Workflow Schedule (scheduled-job, ~configurable cron) — Delayed workflow execution based on schedule
• Task Execution Timeout (async-processing, ~task_timeout config) — Python code execution is killed after timeout
• Graceful Shutdown (async-processing, ~graceful_shutdown_timeout) — Service waits for running tasks before shutdown

Control Points

• AI Features Toggle (env-var) — Controls: Enable/disable AI agent capabilities. Default: N8N_AI_ENABLED
• Personalization Toggle (env-var) — Controls: User personalization features. Default: N8N_PERSONALIZATION_ENABLED
• Task Concurrency (threshold) — Controls: Maximum concurrent Python task execution. Default: max_concurrency
• Security Allowlist (runtime-toggle) — Controls: Which Python modules can be imported. Default: stdlib_allow, external_allow

Package Structure

This monorepo contains 10 packages:

Technology Stack

Key Components

• GlobalConfig (config) — Central configuration management using convict schema with environment overrides packages/cli/src/config/index.ts
• DeduplicationHelper (service) — Service for deduplicating workflow execution data packages/cli/src/deduplication/index.ts
• BinaryDataService (service) — Manages binary file storage and retrieval for workflow data packages/core/src/binary-data/index.ts
• Cipher (utility) — Encryption utilities for securing sensitive workflow data packages/core/src/encryption/index.ts
• BaseError (type-def) — Base error class hierarchy for workflow execution errors packages/workflow/src/errors/index.ts
• Agent (class) — AI agent builder with LangChain integration for conversational workflows packages/@n8n/agents/src/index.ts
• BaseChatModel (class) — Abstract base class for implementing chat model integrations packages/@n8n/ai-node-sdk/src/index.ts
• N8nBinaryLoader (utility) — Utility for loading binary data into AI processing pipelines packages/@n8n/ai-utilities/src/index.ts
• CodeWorkflowBuilder (class) — AI-powered workflow generation from natural language descriptions packages/@n8n/ai-workflow-builder.ee/evaluations/cli/index.ts
• TaskRunnerServer (service) — Python task execution server with security sandboxing packages/@n8n/task-runner-python/src/main.py

Configuration

codecov.yml (yaml)

• codecov.max_report_age (string, unknown) — default: off
• codecov.require_ci_to_pass (boolean, unknown) — default: true
• bundle_analysis.status (string, unknown) — default: warning
• bundle_analysis.warning_threshold (string, unknown) — default: 1%
• comment.require_bundle_changes (string, unknown) — default: bundle_increase
• comment.bundle_change_threshold (string, unknown) — default: 50Kb
• comment.require_changes (string, unknown) — default: coverage_drop OR uncovered_patch
• coverage.status.patch (boolean, unknown) — default: false
• +10 more parameters

cubic.yaml (yaml)

• version (number, unknown) — default: 1
• reviews.enabled (boolean, unknown) — default: true
• reviews.sensitivity (string, unknown) — default: medium
• reviews.incremental_commits (boolean, unknown) — default: true
• reviews.show_ai_feedback_buttons (boolean, unknown) — default: false
• reviews.custom_instructions (string, unknown) — default: ## Step 1: Fetch Current Guidelines 1. Fetch the current [CONTRIBUTING.md](https://github.com/n8n-io/n8n/blob/master/CONTRIBUTING.md) from the repository's main branch 2. Navigate to the "Community PR Guidelines" section 3. Use ONLY this live version - ignore any cached/embedded rules ## Step 2: Review Process Evaluate the PR against the rules in the fetched Community PR Guidelines. ## Step 3: Test Requirement Interpretation BE REASONABLE when evaluating test coverage: **PASS if:** - Core functionality has tests - Critical paths are tested - Coverage is reasonable (not necessarily 100%) **DO NOT require tests for:** - Exports, types, configs - Metadata files - Version files Approve if reasonably tested. Let humans handle edge cases. ## Step 4: Design System Style Rules Follow `.agents/design-system-style-rules.md` for all CSS/SCSS/Vue style review guidance. Enforcement level: - Strong warning: hard-coded visual values, legacy token usage, deprecated style/component surfaces - Soft warning: token-to-token substitutions (ask for intent to avoid accidental regressions)
• reviews.custom_rules (array, unknown) — default: [object Object],[object Object],[object Object],[object Object],[object Object]
• pr_descriptions.generate (boolean, unknown) — default: false
• +2 more parameters

lefthook.yml (yaml)

• pre-commit.commands.biome_check.glob (string, unknown) — default: packages/**/*.{js,ts,json}
• pre-commit.commands.biome_check.run (string, unknown) — default: pnpm biome check --write --no-errors-on-unmatched --files-ignore-unknown=true --colors=off {staged_files}
• pre-commit.commands.biome_check.stage_fixed (boolean, unknown) — default: true
• pre-commit.commands.biome_check.skip (array, unknown) — default: merge,rebase
• pre-commit.commands.prettier_check.glob (string, unknown) — default: packages/**/*.{vue,yml,md,css,scss}
• pre-commit.commands.prettier_check.run (string, unknown) — default: pnpm prettier --write --ignore-unknown --no-error-on-unmatched-pattern {staged_files}
• pre-commit.commands.prettier_check.stage_fixed (boolean, unknown) — default: true
• pre-commit.commands.prettier_check.skip (array, unknown) — default: merge,rebase
• +12 more parameters

pnpm-workspace.yaml (yaml)

• packages (array, unknown) — default: packages/*,packages/@n8n/*,packages/frontend/**,packages/extensions/**,packages/testing/**
• catalogMode (string, unknown) — default: strict
• catalog.@azure/identity (string, unknown) — default: 4.13.0
• catalog.@codemirror/autocomplete (string, unknown) — default: 6.20.0
• catalog.@codemirror/commands (string, unknown) — default: 6.10.1
• catalog.@codemirror/lang-css (string, unknown) — default: 6.3.1
• catalog.@codemirror/lang-javascript (string, unknown) — default: 6.2.4
• catalog.@codemirror/lang-json (string, unknown) — default: 6.0.2
• +109 more parameters

Related Fullstack Repositories

Frequently Asked Questions

Analyzed on March 31, 2026 by CodeSea. Written by Karolina Sarna.