AgentA on Microsoft Store: How Gemma 4 Turned a Local Agent into a Private AI Employee
A follow-up to Building AgentA — from architecture to a published Windows product powered by Google Gemma 4, now available for freelancers and small teams who refuse to upload confidential work to cloud AI.
TL;DR
- AgentA is live on the Microsoft Store as a local-first Windows AI assistant.
- Google Gemma 4 (E4B / 12B) unlocks chat, vision, and audio on ordinary laptops via Ollama.
- Core value: organize files, prioritize email, extract tasks, and answer questions from your documents — privately.
- No per-token cloud LLM fees. 14-day Personal Pro trial included.
From concept to published product
Earlier this year we shared the architecture behind AgentA: a privacy-first desktop agent for files and email, with a local LLM, sandboxed execution, and human control over critical actions.
That was the blueprint. This is the milestone.
AgentA is now published on the Microsoft Store — a production Windows app you can install, try for 14 days with full Personal Pro features, and use as your private AI workspace for documents, email, Telegram, and daily office automation.
The product promise is simple and deliberate:
Your own private AI employee that organizes documents, prioritizes emails, extracts tasks, answers questions from your own knowledge, and automates everyday office work — entirely on your Windows PC, without cloud AI fees or sacrificing data privacy.
Why Gemma 4 changed everything
Local agents only win when the model is good enough for real work and fast enough for everyday use. Google’s Gemma 4 family delivered that leap for AgentA.
In the published app, recommended models are:
- Gemma 4 E4B — best speed for daily chat, vision, and audio (~10 tokens/sec). Recommended default.
- Gemma 4 12B — same multimodal capabilities, stronger extraction quality when you prefer a larger model (~5 tokens/sec).
Both support:
- Chat & tools — natural-language control of files, RAG, invoices, contracts, and urgent tasks
- Vision — understand photos and image documents (Vision LLM, not OCR)
- Audio — Telegram voice notes → local transcription → actionable urgent tasks
That multimodal stack is what turns AgentA from “a clever file helper” into a practical productivity system: one model family that can read a scanned invoice PDF via OCR, interpret a screenshot via vision, and convert a voice note into a task — all on your machine through Ollama.
What the published app delivers today
Document intelligence you feel immediately
- Find any contract or invoice in seconds — even across thousands of files
- Search by meaning, not filename (local RAG over your private data)
- Scan & index folders without moving files (non-destructive by default)
- Summarize long PDFs, Word, Excel, images, and web articles into TLDR + key points
- Detect duplicates and reduce archive clutter
- Document Library with invoices/contracts registers and custom type gates
Email and Telegram that respect your control
- Optional IMAP watcher with AI priority scoring and categorization
- Reply drafts you review and send manually — never auto-send
- Telegram bot (whitelist only): files, links, voice → structured summaries and urgent tasks
- Daily leads from screenshots and urgent-action digests
Privacy, cost, and hardware that fit real professionals
- Core AI inference stays on your PC — no cloud document upload for model processing
- No per-token OpenAI-style API fees — Ollama + open models you control
- Runs on ordinary Windows laptops/desktops — no dedicated GPU required
- Point-and-click UI for non-technical users; dark and light themes
- 14-day Personal Pro trial; monthly or lifetime upgrade via Microsoft Store IAP
Who this is for
AgentA is built for freelancers, consultants, and small teams who live in documents and inboxes — and who cannot accept “upload everything to the cloud” as the price of AI productivity.
- Professionals handling confidential contracts, invoices, and client files
- Solo operators who need speed without hiring admin support
- Anyone who wants AI as a specialist assistant — with guardrails and human approval for critical actions
The bigger story
In our earlier manifesto on local SLM agents, we argued that affordable, private, human-supervised AI must reach everyday users — not only enterprises with cloud budgets.
AgentA on the Microsoft Store is that thesis shipped as a product: local models, real workflows, store distribution, and a clear upgrade path — without forcing people to rent intelligence by the token.
Gemma 4 made the experience sharp enough to publish with confidence. Microsoft Store makes it easy enough to try.
Get AgentA for Windows
Install from the Microsoft Store, connect Ollama with Gemma 4, and start indexing the folders you already use.
