Ghost PM

Inspiration

Product managers spend a huge amount of time turning messy chat messages into structured tasks, deadlines, and follow-ups. We wanted to automate that repetitive work so PMs can focus on strategy and decision-making instead of manual task admin.

What it does

Ghost PM turns chat messages into actionable PM tasks in Notion.

From a Telegram PM group, it:

Detects task-like messages Extracts and normalizes task title, priority, and due date Asks for missing due dates when needed Prevents duplicate task creation For research-oriented tasks, runs web research and stores a concise summary + links Saves everything to a structured Notion database

How we built it

We built Ghost PM using:

OpenClaw for agent orchestration and chat workflow handling Notion API for structured task storage and status management OpenAI for coherent research summarization Tavily for web research enrichment on research-type tasks The system uses deterministic parsing + guardrails (dedupe keys, confidence thresholds, due-date validation) to keep behavior reliable.

Challenges we ran into

Handling edge cases in natural-language date parsing (e.g., “tomorrow”, “next Friday”, “in 3 days”) Mapping Notion field types correctly (e.g., URL vs rich text fields) Balancing autonomy with safety (when to auto-create vs ask clarification) Deployment friction: repository sharing and workflow from a VPS environment

What’s next for Ghost PM

Smarter autonomy policies and personalized PM preferences Better project/context grouping in Notion Reminder workflows and proactive follow-ups Improved research quality with source ranking and citation formatting Multi-channel support beyond Telegram Team-level collaboration features (assignees, ownership, status nudges)

Built With

• notion
• openai
• openclaw
• python
• tavily