Membrow

Membrow is a lightweight, developer-focused desktop browser application designed for rapid research capture and memory synthesis. It integrates directly with Memron via the Model Context Protocol (MCP) to clip, extract, and ground developer intelligence (papers, tools, models, agents, and skills) into persistent memory buckets.

Key Features

• Actual Desktop Browser: Runs natively via Electron with an embedded Chromium guest web engine (<webview>) to browse any site (X/Twitter, GitHub, ArXiv, Hugging Face, Reddit, Google) without X-Frame-Options or CORS restrictions.
• Top-Corner Snapshot & Clip: Compact, minimalist one-click button to capture high-res page snapshots and automatically extract structured metadata (Title, Platform, Author, Category, Takeaways, Tags).
• 5 Developer Knowledge Categories:
  • Research: Academic preprints, architectures, ablation studies, and benchmarks.
  • Tools: Developer utilities, CLIs, libraries, and open-source packages.
  • Agents: Autonomous cognitive frameworks, multi-agent orchestrators, and workflows.
  • Models: Open weights, LLMs, vision models, quantizations, and checkpoints.
  • Skills: Antigravity skills, function schemas, and system prompt specifications.
• Hamburger Sidebar Drawer:
  • Memron MCP Connection: Connect to Memron MCP servers via standard JSON-RPC protocol (tools/call for store_memory, search_memory, list_buckets) with live connection badges and latency counters.
  • Knowledge Data Bucket: Full searchable knowledge feed with category filters, Markdown/JSON export, and tab integration.
• Anti-AI Minimalist Aesthetic:
  • Dark slate monochromatic palette with high-contrast precision typography.
  • Strictly zero emojis across all interfaces.
  • Crisp SVG vector iconography from @lucide/svelte.
• Omnibox & Tab Strip: Multi-tab browsing, direct URL navigation, DuckDuckGo search integration, and full history controls (Back, Forward, Reload, Home).

Architecture

membrow/
├─ apps/
│  ├─ web/                       # Desktop Browser Shell (Svelte 5 + Vite + Electron)
│  │  ├─ electron/               # Native Electron main process & IPC preload bridge
│  │  │  ├─ main.cjs             # Window management, webview guest host, native capturePage()
│  │  │  └─ preload.cjs          # Typed membrowDesktop bridge
│  │  └─ src/
│  │     ├─ lib/components/      # BrowserChrome, TabBar, WebViewContainer, SidebarDrawer, SnapshotModal
│  │     ├─ lib/mcpClient.ts     # Memron MCP Client & Persistent Knowledge Bucket
│  │     ├─ lib/extractor.ts     # Auto-classification & metadata extraction pipeline
│  │     └─ routes/+page.svelte  # Unified browser shell
│  └─ api/                       # Go + Fiber backend (optional worker service)
├─ docs/                         # Architecture, workflow, and error handling reference
└─ scripts/
   ├─ desktop.ps1                # One-click desktop launcher (PowerShell)
   └─ dev.ps1                    # Multi-process dev helper

API endpoints

• GET /health
• POST /search (passthrough provider search)
• POST /agent/run (starts full workflow asynchronously)
• GET /agent/run/:id (returns run status, trace, and result/error)

Environment variables

API (apps/api/.env.example)

Variable	Default	Purpose
API_PORT	8080	Fiber listen port
LOG_LEVEL	info	Structured log level
REQUEST_TIMEOUT_SECONDS	15	Request-level timeout
AGENT_RUN_TIMEOUT_SECONDS	45	Max duration per agent run
SHUTDOWN_TIMEOUT_SECONDS	10	Graceful shutdown timeout
DEFAULT_SEARCH_LIMIT	5	Default search result count
SQLITE_PATH	./data/membrow.db	SQLite file path
SEARCH_PROVIDER	mock	mock or serpapi
SERPAPI_KEY	(empty)	Optional real search provider key
WORKER_POOL_SIZE	4	Fetch/extract concurrency limit
MAX_RETRIES	2	Retry count for retryable calls
RETRY_BACKOFF_MS	250	Base retry backoff
FETCH_USER_AGENT	membrow-bot/0.1	HTTP fetch user-agent

Web (apps/web/.env.example)

Variable	Default	Purpose
PUBLIC_API_BASE_URL	http://localhost:8080	Base URL for backend API

Local setup

Prerequisites

• Go 1.22+
• Node.js 22+
• npm 10+

Windows (PowerShell)

Copy-Item .\apps\api\.env.example .\apps\api\.env
Copy-Item .\apps\web\.env.example .\apps\web\.env
cd .\apps\api; go mod tidy
cd ..\web; npm install
cd ..\..
.\scripts\dev.ps1

macOS/Linux

cp ./apps/api/.env.example ./apps/api/.env
cp ./apps/web/.env.example ./apps/web/.env
cd ./apps/api && go mod tidy
cd ../web && npm install
cd ../..
./scripts/dev.sh

Docker development

docker compose up --build

• Web: http://localhost:5173
• API: http://localhost:8080

Testing

API tests

cd apps/api
go test ./...

Web checks

cd apps/web
npm run lint

Scaling roadmap

1. SQLite → Postgres
   • keep RunStore interface, add Postgres-backed implementation
2. Redis cache + queueing
   • cache repeated queries and stage intermediate artifacts
   • push fetch/extract work onto queue workers
3. Distributed workers
   • move stage execution from in-process goroutines to horizontally scalable workers
4. Run orchestration hardening
   • add dead-letter queues, retry budgets, and per-provider circuit breaking
5. Observability expansion
   • OpenTelemetry traces and metrics exporters

Follow-up TODOs

• Add authentication/authorization and user-scoped run history.
• Add richer extract pipeline (readability heuristics + content quality scoring).
• Add streaming run updates (SSE/WebSocket) instead of polling.
• Add Postgres implementation and migration tooling.
• Add integration tests using mocked external providers.