AI personal assistant setup

{{whyLabel}}: To build an effective assistant, you must first understand where your data currently lives and where the friction points are.

{{howLabel}}:

• List all current tools (e.g., Google Calendar, Todoist, Notion, Outlook).
• Identify 'Dead Zones': Tasks that currently require manual entry (e.g., extracting dates from emails).
• Document your 'Golden Rules': Fixed constraints like 'No meetings before 10 AM' or 'Always 15 min buffer between calls'.

{{doneWhenLabel}}: You have a written list of 3+ specific manual workflows you want the AI to automate.

{{whyLabel}}: A clear persona ensures consistent AI behavior and better instruction following.

{{howLabel}}:

• Use the ROC-T-TOC formula: Role (Executive Assistant), Objective (Manage time & tasks), Context (Professional/Personal balance), Tools (Calendar/Todoist), Tasks (Scheduling/Prioritizing), Operating Guidelines (Tone/Style), Constraints (Privacy/Boundaries).
• Decide on the 'Autonomy Level': Should it ask for permission before every calendar entry or act autonomously?

{{doneWhenLabel}}: A 1-page 'Assistant Charter' defining the AI's personality and rules.

{{whyLabel}}: The LLM is the 'brain' that processes natural language into structured data.

{{howLabel}}:

• For high reasoning: Use OpenAI GPT-4o or Claude 3.5 Sonnet via API.
• For privacy/local: Use Ollama with Llama 3.1 (requires a powerful PC).
• Obtain your API Key and set a monthly usage limit (e.g., $5) to prevent unexpected costs.

{{doneWhenLabel}}: You have an active API key from your chosen provider.

{{whyLabel}}: These platforms act as the 'nervous system,' moving data between the AI and your apps.

{{howLabel}}:

• Choose n8n (Open Source/Self-hosted) for maximum privacy and complex logic.
• Choose Make.com (Cloud-based) for a faster, visual setup with 1,500+ pre-built integrations.
• Create a new 'Scenario' or 'Workflow' and name it 'AI Assistant Core'.

{{doneWhenLabel}}: An account is created and a blank workflow is initialized.

{{whyLabel}}: The AI needs read/write access to your actual schedule to be useful.

{{howLabel}}:

• In your Automation Hub, add a 'Google Calendar' or 'Outlook' node.
• Add a 'Todoist' or 'Notion' node.
• Authenticate both using OAuth2 to ensure secure, token-based access without sharing passwords.

{{doneWhenLabel}}: Successful 'Connection Test' in the automation hub for both tools.

{{whyLabel}}: This tells the AI exactly how to interpret your messy human requests.

{{howLabel}}:

• Input this core instruction: 'You are a scheduling expert. Extract: 1. Event Title, 2. Start Time (ISO8601), 3. Duration, 4. Priority. If time is vague (e.g., "tomorrow morning"), use 9:00 AM. Output ONLY JSON.'
• Add your personal constraints (e.g., 'Never schedule over lunch 12-1 PM').

{{doneWhenLabel}}: The LLM node in your workflow consistently outputs valid JSON from text inputs.

{{whyLabel}}: This allows the AI to actually trigger the calendar/task nodes based on its reasoning.

{{howLabel}}:

• Set up a 'Router' or 'Switch' node in your hub.
• If the AI output contains 'action: create_event', route to the Calendar node.
• If it contains 'action: create_task', route to the Todoist node.
• Map the JSON fields from the AI directly to the tool's input fields.

{{doneWhenLabel}}: A test message like 'Remind me to buy milk tomorrow at 5pm' successfully creates a task.

{{whyLabel}}: Telegram provides a free, secure, and easy-to-use interface for both text and voice.

{{howLabel}}:

• Message @BotFather on Telegram to create a new bot and get an API Token.
• Add a 'Telegram Trigger' node to your automation hub.
• Connect the bot so every message you send to it triggers your AI workflow.

{{doneWhenLabel}}: You can send a message to your bot and see it arrive in your automation hub.

{{whyLabel}}: Voice is the fastest way to capture tasks while driving or walking.

{{howLabel}}:

• Add a logic branch for 'Audio' messages in your hub.
• Send the audio file to the OpenAI Whisper API (model: whisper-1).
• Pass the resulting text into your existing AI scheduling logic.

{{doneWhenLabel}}: Sending a voice note to the bot creates a calendar event or task.

{{whyLabel}}: New systems often fail on edge cases (e.g., time zone shifts or overlapping events).

{{howLabel}}:

• For 7 days, use the assistant for every task but keep your manual system as a backup.
• Log every 'Hallucination' (e.g., AI scheduled something on the wrong day).
• Do not delete the logs; they are essential for prompt tuning.

{{doneWhenLabel}}: 7 days of logs completed with at least 20 interactions recorded.

{{whyLabel}}: To stop the AI from making the same mistakes twice.

{{howLabel}}:

• Update your System Prompt with 'Negative Constraints' (e.g., 'NEVER schedule tasks on Sundays').
• Add 'Few-Shot Examples': Provide 3-5 examples of messy inputs and the perfect JSON outputs you expect.
• Adjust the 'Temperature' setting (set to 0.0 for scheduling to ensure consistency).

{{doneWhenLabel}}: Updated System Prompt deployed in the automation hub.