MailPilot AI
Autonomous Gmail Assistant with LangGraph State Machines, Real-Time Streaming & Human-in-the-Loop Protection
LangGraph StateGraph
Agent Framework
Gemini 2.0 Flash
LLM Engine
11 Gmail Tools
Autonomous Tools
HITL Guardrails
Safety Layer
System Narrative
Engineering Context & Problem Scope
MailPilot AI is an autonomous email agent that empowers users to search, summarize, triage, draft, and reply to Gmail emails using natural language without opening the standard Gmail client. Built with LangGraph and Google Gemini 2.0, it features an enterprise-grade Human-in-the-Loop (HITL) architecture that automatically pauses execution and awaits user review before performing sensitive actions like sending, replying, or deleting messages. Engineered with a three-tier persistence model combining SqliteSaver for graph execution state and SQLModel for encrypted authentication, MailPilot AI streams reasoning states and tokens in real-time over Server-Sent Events (SSE) into a responsive Next.js 16 and React 19 interface.
Key Business & Engineering Outcomes
- Built an autonomous Gmail agent using LangGraph and Google Gemini 2.0 with 11 custom tools for email search, summarization, drafting, and triaging.
- Architected a Human-in-the-Loop (HITL) safety mechanism with SQLite state checkpointing to require manual user approval before sending or deleting emails.
- Developed a full-stack real-time interface using Next.js 16, Tailwind CSS, and FastAPI SSE streaming for token-by-token response rendering.
Integrated Technologies
Architecture Blueprint
Full-Stack System Layers
Architected for high throughput, data consistency, and robust fault-tolerance.
Frontend Layer
- Next.js 16 (App Router)
- React 19 & TypeScript
- Tailwind CSS with Futuristic Dark Theme
- Real-Time SSE Token & Reasoning Stream Handler
- Interactive HITL Approval Cards & Thread Modal
Backend & APIs
- FastAPI Async Server (Python 3.12+)
- LangGraph ReAct StateGraph State Machine
- Custom SSE Streaming Pipeline with Token Deduplication
- Google OAuth 2.0 Token Lifecycle & Session Manager
Data & Persistence
- SqliteSaver Checkpointer (Agent State & Thread Snapshots)
- SQLModel Database (User Profiles & Encrypted Auth Tokens)
- Client LocalStorage (Multi-Session Short-Term Context)
AI & Vector Layer
- Google Gemini 2.0 Flash Lite (langchain-google-genai)
- 11 Autonomous Gmail Tools (Search, Triage, Draft, Send)
- Human-in-the-Loop Interrupt & Deterministic State Resumption
- Domain Guardrails & Anti-Hallucination Scope Filters
Capabilities & Features
Core System Features
Key architectural capabilities and modular components engineered for this platform.
11 Autonomous Gmail Tools
Comprehensive email tool suite including search_emails, read_email, get_thread, archive_email, mark_as_read/unread, add_label, create_draft, and protected send/reply/delete operations.
Human-in-the-Loop (HITL) Safety Guardrails
Automatically interrupts execution on sensitive/external actions (send, reply, delete), rendering interactive approval cards with full recipient, subject, and body previews before resuming seamlessly from state snapshots.
Three-Tier Persistence Architecture
LangGraph SQLite checkpointer (SqliteSaver) preserves graph thread snapshots, SQLModel manages authenticated user profiles and encrypted OAuth tokens, and local cache preserves multi-session history.
Real-Time Token & Reasoning Streaming
Server-Sent Events (SSE) stream model tokens, animated thinking indicators, and deduplicated tool usage metadata live to the client.
Responsive Modern UI & Thread Inspection
Built with Next.js 16 and Tailwind CSS featuring quick-action capability cards, email thread inspection modal (with HTML/plain text view), and off-canvas mobile drawer navigation.
Interested in Building a Similar System?
Let's discuss architecture blueprints, tech stack selection, and production deployment tailored for your requirements.