Files

2.5 KiB

PAI Roadmap

Personal AI OS - automate your digital life with natural language.

Completed

  • Foundation - CLI, SQLite database, Pydantic models, configuration
  • Intent Engine - Natural language → structured automation specs
  • Gmail Connector - OAuth, search, labels, entity extraction
  • Execution Engine - Run automations, dry-run mode, history
  • Trigger System - Email watcher with condition matching
  • MCP Architecture - Model Context Protocol for all connectors (pai_mcp/)
  • Outlook Connector - Email + calendar via Microsoft Graph (external MCP)

Short-Term

Priority: Enable the 3 MVP workflows from the spec.

Feature Purpose
Google Sheets MCP Log extracted data to spreadsheets (pai_mcp/sheets.py)
Google Drive MCP Save attachments to organized folders (pai_mcp/drive.py)
Attachment Executor Download and save email attachments
Schedule Triggers Run automations on cron schedules
Approval Gates Require confirmation for high-stakes actions

Medium-Term

Priority: Polish, learn, expand.

Feature Purpose
Learn Stage Capture corrections, auto-refine automations
Semantic Triggers LLM-powered condition matching ("invoice-related")
More MCP Servers Notion, Slack, Google Calendar
Cost Tracking Monitor API usage and LLM costs
Error Recovery Retry logic, partial rollback support

Long-Term

Priority: Advanced capabilities and scale.

Feature Purpose
Local Semantic Index Embeddings for cross-source queries
Code Sandbox Safe execution of LLM-generated code
Sync Layer CRDT-based state across devices
Visual Canvas Drag-drop workflow builder UI
Community Library Share and fork automations
Mobile Companion Approve actions on the go

MVP Workflows

These 3 workflows drive prioritization:

  1. Client Email Triage - Categorize emails, extract deadlines (working)
  2. Invoice Processing - PDF → folder → spreadsheet (needs Drive + Sheets)
  3. Weekly Client Report - Aggregate activity per client (needs Schedule)

Architecture Principles

  • Local-first - Your data stays on your machine
  • CLI-first - Core logic before UI
  • Flat structure - Easy to navigate and extend
  • Async throughout - Ready for I/O-heavy workloads
  • MCP-based - All connectors are MCP servers in pai_mcp/
  • Pluggable - Connectors and LLM providers are extensible