All projectsPROJECT/04
AI for Bharat Hackathon
FlowSync
AI Project Memory for Coding Agents
Gives AI coding agents persistent memory across sessions via the Model Context Protocol. Agents log why they made decisions and query weeks of project history; git pushes auto-capture context as a fallback, all through a serverless AWS Bedrock pipeline.
FlowSync · SchematicFIG/04
The Problem
AI coding agents are stateless. Every session starts from zero with no memory of past decisions or architecture. FlowSync gives agents persistent project memory via the Model Context Protocol.
How It Works
VS Code Extension + MCP Server (TypeScript)→
API Gateway→
AWS Lambda (Python 3.12)→
Amazon Bedrock (Nova Pro + Nova Lite) + Titan Embeddings→
DynamoDB + S3→
Next.js Team Dashboard
Key Features
- 5 MCP tools discovered natively by Copilot, Cursor, and Claude Code
- log_context records the why behind each change: decisions, risks, and reasoning per commit
- search_context answers natural-language questions with strict source citations back to commits
- Git-push auto-capture fallback, so the project brain grows even for developers without an agent
- Titan embeddings stored beside context items in DynamoDB; no separate vector database to operate
- Branch-aware RAG plus a real-time team dashboard of decisions, risks, and pending tasks
- Shipped as a VS Code extension with scrypt-hashed project tokens and a 1-hour RAG response cache
Technical Highlights
- Event-driven AWS pipeline: raw events land in DynamoDB in about 200ms, Bedrock extraction runs asynchronously, and branch-aware RAG answers queries at ~1.2s median latency.
- Tiered model routing sends once-per-commit extraction to Nova Pro and interactive chat to Nova Lite, cutting inference cost by roughly 60%.