AI Agent Workflow

Sanitized n8n workflow for a personal AI operations agent.

This repo is a public, recruiter-friendly version of a real automation workflow. It keeps the architecture and implementation shape visible while removing private credentials, personal account IDs, private repository names, webhook secrets, and local paths.

What this shows

• Telegram and n8n chat input for a personal assistant workflow
• Voice transcription and photo analysis branches
• AI agent loop with tool routing
• Gmail search/read/draft actions
• Google Calendar read/create actions
• Notion CRM search/read/create/update actions
• GitHub-backed markdown memory as a portable memory store
• Scheduled morning, pre-meeting, midday, evening, and weekly heartbeat prompts
• Validation script that blocks leaked credentials before publishing

Architecture

Telegram / Chat / Schedule
        |
Normalize input and route by source
        |
AI Agent
        |
Tool router
        |
Gmail | Calendar | Notion | GitHub memory
        |
Final response back to chat or Telegram

Repo structure

workflows/ai-agent-workflow.sanitized.json  Sanitized n8n export
scripts/validate-workflow.mjs              JSON and secret-leak validation
.env.example                               Example environment variables
SECURITY.md                                Handling rules for secrets

Import

1. Import workflows/ai-agent-workflow.sanitized.json into n8n.
2. Reconnect credentials for Telegram, OpenAI, DeepSeek, Gmail, Google Calendar, Notion, and GitHub.
3. Replace placeholder values such as YOUR_TELEGRAM_CHAT_ID, YOUR_GITHUB_USER/YOUR_MEMORY_REPO, work@example.com, and personal@example.com.
4. Run a dry test with chat input before enabling schedules or Telegram triggers.

Validate before publishing

npm test

The validator fails if the workflow contains n8n credential references, private identifiers from the original workflow, common token formats, or real secret-like values.

Notes

This is a showcase workflow, not a drop-in production bundle. The original private deployment used real credentials and personal memory paths. Those have intentionally been replaced with placeholders.