Flow Guardian | Devpost

Inspiration

AI coding assistants lose context when the session ends.

Architectural decisions, debugging history, and learnings disappear. For teams, knowledge fragments across individual chats, causing repeated mistakes and wasted time.

Flow Guardian fixes this by giving AI a persistent, shared memory layer.

What Flow Guardian Does

Flow Guardian is a background daemon that continuously captures and serves context across all Claude Code sessions.

It works as:

A local daemon tracking every Claude session
An MCP server AI tools can query directly
A web + API layer for team access and automation

Highlights

Always-on daemon Passively monitors all Claude Code sessions and extracts decisions, blockers, and learnings—no prompts required.
Persistent memory Context survives across sessions, machines, and teammates.
MCP-native integration Any MCP-compatible AI (Claude Code today, others tomorrow) can recall and write memory.
API + Web client Memory is accessible programmatically and via a web UI for search, review, and docs.
Team-shared knowledge Learnings sync automatically via Backboard.io.
Automatic issue creation Detects blockers in conversations and creates Linear issues with full context.
Fast local-first search Instant local recall, cloud fallback only when needed.

How We Built It

Flow Guardian Daemon (Python) Session watcher + context extractor
MCP Server Tools: flow_recall, flow_capture, flow_learn, linear_create_issue
Remote MCP Transport HTTP + SSE for multi-user, remote access
Cerebras (zai-glm-4.7) Ultra-fast classification and routing
Backboard.io Cloud-synced team memory
Linear GraphQL API Context-aware issue automation
Next.js Web Client Memory search, activity feed, docs