Перейти к содержимому

magpie: The Local-First AI Work Journal & Developer Posting Desk (100% Offline)

Warisha Bilal

0:00 / 0:00

magpie: The Local-First AI Work Journal & Developer Posting Desk (100% Offline)

6 просмотров · 10 дней назад
Warisha Bilal
1 подписчик
6 просмотров · 10 дней назад
Turn scattered git commits, AI chats, and raw notes into source-linked journals, verified posts, and evidence-grounded resume bullets — running 100% locally on your machine. In this demo, we walkthrough Content Desk (cdesk), a local-first assistant built for developers and researchers who work across multiple AI tools, git repos, and social platforms. ================================================== ✨ WHAT MAKES CONTENT DESK DIFFERENT? ================================================== Most "AI caption generators" start with a blank prompt and hallucinate facts. Content Desk starts with the work you actually did. • 🔒 100% Private & Offline: Powered by local open-weight models via Ollama and local SQLite. Zero cloud API keys required, zero telemetry — nothing ever leaves your laptop. • 🧾 Receipts & Deterministic Verification: The LLM writes, but deterministic Python code verifies. Every single journal line cites its source capture, and strict verifiers catch invented dates, numbers, links, and handles before saving. • 📥 Frictionless Capture Inbox: Drop markdown files, import ChatGPT / Claude chat exports, or scan daily git commits with a single CLI command (`cdesk capture`). • 🎙️ Workspaces & Voice Matching: Separate channels for LinkedIn, Substack, technical societies, and personal blogs. Enforces custom banned-phrase lists, platform constraints, and style embeddings. • 📅 Planner & Cadence Nudges: Week and month view, draft reminders, .ics calendar sync, and "you haven't posted in N days" nudges. • 💼 Job Tracker & Resume Tailoring: Match job descriptions against real journal evidence to generate verifiable resume bullets and cover letters grounded in facts. • 📦 Context Packs: Instantly bundle recent decisions, journal lines, and open questions into a compact brief to paste into any AI chat. ================================================== ⏱️ TIMESTAMPS ================================================== 0:00 - Introduction & The Problem with Fluffy AI Desks 0:35 - Frictionless Capture: Git Commits & AI Chat Exports 1:15 - The Work Journal: Deterministic Verification & Citations 2:00 - Multi-Workspace Drafting & Voice Matching (LinkedIn vs Substack) 2:50 - Platform Rule Checks & Fact-Checking Hallucinations 3:35 - Content Planner, Reminders & .ics Calendar Export 4:15 - Ideas Organizer & Duplicate Detection 5:00 - Job Tracker & Evidence-Based Resume Tailoring 5:45 - Wrap-up & Local Architecture ================================================== 🛠️ TECH STACK ================================================== • Orchestration: LangGraph & deterministic verify nodes • Local Models: Ollama (Qwen / Llama / Mistral) with template fallback • Frontend: Streamlit • Storage: SQLite + FTS5 full-text search • CLI: Typer & Rich 🔗 Repository: https://github.com/Warisha-Bilal7/magpie #LocalFirst #Ollama #LangGraph #Streamlit #OpenSource #AIDeveloper #Python #PrivacyFirst #DevTools