AgentA is your private AI employee.

Versatile Chat RAG on your Windows PC: Librarian · Analyze · Editor. Index your files. Ask. Instruct. Confirm. Your storage. Your privacy. Email triage — prioritize and draft. Urgent tasks — capture what matters before it slips. Daily leads — turn screenshots and messages into a simple pipeline. Registers — invoices and key document types, one click away. Telegram — ask your agent and track tasks from your phone. No cloud AI bills. Your files stay private.

Get it from Microsoft Watch live demo See what it solves AgentA Max · Mac
Scroll

See it work

Real AgentA run on Windows — timed on an everyday laptop.

Screen recording of Local RAG Q&A with Gemma 4 E4B Q4 — three logistics questions answered from your own files. Click play.

Demo 01 · Local RAG · Chat

Three logistics questions — answered from your own files

Local RAG searches your project documents first, then answers from those results — not the open web. Docs: AgentA Logistics demo pack (contracts · B/Ls · certificates). Mode: Chat · project knowledge (local files only). Same Chat also supports folder Scope, numbered multi-ask, foldable Sources, and OCR-aware citations after Scan & Index.

  1. Find the seller on the Copper Finch purchase contract
  2. Find the port of discharge on the Summit Trail bill of lading
  3. List SKUs on the Silver Heron packing list

Timing: total ~26 sec · LLM generation 96 tokens · 4.8 tok/s (~20 sec)
Model: Gemma 4 E4B Q4 · Hardware: Lenovo IdeaPad Slim 5 16IAH8 · Core i5 12th Gen · 16 GB DDR5

Key benefits

The daily pains. The clear relief.

Freelancers, consultants, and small businesses lose hours every week to lost invoices, buried contracts, messy folders, and overflowing email — then pay again for cloud AI that sends private work off their machine.

AgentA is the AI employee that stays on your Windows PC: it indexes your documents for versatile Chat RAG (ask, edit, export), keeps your files private where you already store them, and runs local AI on your computer instead of always hitting giant cloud data centers — so routine work uses less shared energy and puts a lighter load on the environment. (Why local AI access matters)

Today’s local language models have reached a practical new level: with Google’s Gemma 4 family, high-quality answers for everyday office work are achievable on ordinary mass-market laptops — not only on specialist AI workstations. AgentA is built around that shift: private AI that fits the hardware people already own.

AgentA is not built to fire or replace people. It is built to make everyday work easier — take over the routine, and give people back free time for real work and real life.

The product is shaped by 20+ years of professional experience in business execution — not by a “try everything” automation mindset. AgentA is a pure desktop, point-and-click app: set it up once, then simply use it. No coding skills, no PowerShell, and no terminal work are required. Meaningful actions stay under your approval, inside clear guardrails — so there are no unintended deletions or other unpleasant surprises. You stay in control; the agent assists.

01

Never lose an invoice again.

02

Find any contract in seconds — even if you forgot its name.

03

Your emails sorted before your morning coffee.

04

Private AI that never uploads your confidential documents.

05

Index your archive into a private knowledge library — without rearranging your folders.

06

Three Chat modes: Librarian (ask indexed content), Analyze (LLM review / compare), Editor (instruct AgentA, review reworked text, confirm save to .docx or PDF).

07

Stop searching folders. Start finding answers instantly.

08

Transform long reports and PDFs into clear summaries in seconds.

09

Capture every important task from email, Telegram, and voice — before it is forgotten.

10

Enjoy AI power without monthly API bills or losing control of your data.

11

Use AI on your PC instead of always calling cloud data centers — lighter energy use, gentler on the planet.

12

Not made to replace people — made to ease their day and free time from routine tasks.

Why people hire this AI employee

Works where you work

Runs on your PC today — Windows 10 & 11. Ordinary laptops welcome (DDR4 minimum; DDR5 or newer recommended). AgentA Max for Mac (MLX) is coming — see the Max section.

Gemma 4–ready quality

Modern local LLMs — led by Google’s Gemma 4 family — bring strong, usable results to everyday hardware, so AgentA can deliver serious help without a dedicated GPU farm.

Kills admin drag

Less searching, less re-reading, fewer forgotten follow-ups — more billable time.

Grows with your archive

One local workspace instead of juggling single-purpose tools and messy folders.

Predictable cost

No per-token cloud fees. Optional Personal Pro monthly or lifetime via Microsoft Store.

Librarian · Analyze · Editor

Ask indexed content (Librarian), deepen review or compare scoped files (Analyze), or instruct updates and confirm a new .docx or PDF (Editor) — plus optional live web when you need public, up-to-date info.

Lighter footprint

Local AI on your machine means less routine traffic to power-hungry cloud data centers — better for energy use and the planet.

People first

Not a tool to push people out. A helper that clears routine work so you get easier days and more free time.

Point-and-click

A pure desktop app — pick folders, scan, chat. No coding, no PowerShell, no terminal. Tune once, then just use it.

You stay the boss

Actions that matter need your approval. Rename and move stay off until you turn them on. Built on solid guardrails — no silent file surprises.

Business-proven design

Designed from 20+ years of real business-execution practice — calm, practical, and safe for everyday professionals.

Safe to trust — built to assist, not to take over

Agent apps can feel risky. AgentA is intentionally different: a familiar Windows desktop experience, clear permissions, and human approval for critical steps. Configure once, work with confidence — without scripts, without surprise deletions, and without giving up control of your files.

AgentA helps people — it does not replace them — it clears routine office work so your team gets time back, on your Windows PC.

The idea

Three moves. That’s the whole product.

Your AI employee sits between your everyday files and a clear result — indexing, understanding, and acting on your machine.

01
Your stuff

Folders, PDFs, photos, mail, Telegram — already on your machine.

02
Your AI employee

AgentA indexes, understands, and chats over your files — locally on Windows. You keep your own storage layout.

03
You get results

Answers, tasks, registers, summaries — ready to use.

Function map Your AI employee at a glance
Files Scan & Index · Process
Email Priority · drafts · search
Telegram Text · files · voice → tasks
feeds into
AI Employee Local RAG · LLM · Vision · OCR · Live web (DuckDuckGo)
delivers
Chat Ask your own data
Library Catalogue of files
Documents Invoices · contracts
Urgent Tasks & leads
plus
Settings Models · folders · privacy
Deal Map Business tier

Inside the app

See AgentA on your screen.

Swipe or use the arrows — real windows from Dashboard through Chat, Registers, and Settings.

AgentA Personal Pro upgrade plans
01 Personal Pro — monthly or lifetime
AgentA Dashboard overview
01a Dashboard — LLM and watcher status at a glance
AgentA Files scan and RAG index
02 Files — Scan & Index, RAG, duplicates, rename by pattern
AgentA Document Library
03 Document Library — indexed archive, OCR quality scores, Re-index RAG
AgentA Emails and Messages
04 Emails & Messages — priority inbox and Telegram
AgentA Registers invoices and contracts
05 Registers — invoices, contracts, type gates
AgentA Chat assistant
06 Chat — ask indexed files, edit text, save .docx, multi-chat
AgentA Logs diagnostics
07 Logs — diagnostics when you need them
AgentA Settings
08 Settings — plan, folders, dark mode

Tip: on phone, swipe left or right across the screenshot.

Chat capability

Three modes in AgentA Chat.

After you Scan & Index your files, Chat works in three clear modes — all local, all over your private corpus. Switch intent with Scope, the Analyze toggle, or revise-and-export language.

01 · Librarian

Ask any question about the content of your indexed documents. Fast, grounded answers from your private index — with cited Sources — when Analyze with LLM is off (default Librarian mode).

02 · Analyze (LLM)

Scope an indexed file and request deeper local-LLM analysis: improve or review text, or compare documents against the parameters you specify. Turn Analyze with LLM on (or use Compare scope) for fuller, slower analysis.

03 · Editor

Instruct AgentA in plain English to update a found or Scoped document — a clause, passage, or other wording. Review the reworked text AgentA proposes (Before/After), then confirm to save a new .docx or PDF. You do not edit the document in Chat; you direct and approve. Original file untouched.

How they fit together: Librarian answers “what does this say?” · Analyze digs into quality, comparison, and improvement suggestions · Editor applies your instructions, shows reworked text for your review, and exports a new file only after you confirm. Commands and examples: Chat & Telegram commands.

Audience

Built for people who live in folders.

Freelancers who want billable hours back Consultants with confidential client files Small businesses drowning in admin Teams who want help — not replacements Messy Downloads, real deadlines Privacy-first professionals No coding required Windows 10 and 11 AgentA Max · Mac — coming soon
Coming soon — AgentA Max for Mac

A native Apple Silicon edition is in development: local MLX inference, unlimited My Agents, Accept-gated mail & social, Drive connectors, IDE, and more. Explore the AgentA Max section or read the Max blog post. Windows remains available today on the Microsoft Store.

For organizations

AgentA for Business

Beyond Personal on the Microsoft Store — a private AI workspace for small teams, freelancers with client archives, and companies that need their own data, Deal Map, and server-side workflows under their control.

Fine-tuned on your data

We can personalize AgentA by fine-tuning models on your organization’s materials — so answers, classifications, and routines reflect your language, documents, and operating practice, not a generic public model alone.

On your company server

AgentA can be installed in your environment to process company email and file flows under your policies — keeping sensitive work inside your perimeter while people still get a clear, point-and-click experience.

Deal Map visualization

See each deal as a living map: project folder at the center, contract anchored, linked documents around it — so teams stop hunting files and start seeing the deal structure at a glance.

Local knowledge (RAG)

Index server-side files and mail into a private knowledge base. Chat supports folder/file/email scope, numbered multi-ask, foldable Sources with page citations, Document Library OCR quality scores, multi-chat sessions, and Editor / Revise & Export — instruct AgentA, review the proposed text, confirm, and save a new .docx or PDF — grounded answers and assisted edits from your real documents, not the public web by default.

Remote access to your agent

Authorized users can reach AgentA on the company server through remote tools and ask questions about shared files from where they work — same private knowledge, without copying archives onto every laptop.

Built for small teams & freelancers

Sized for lean operations: freelancers with confidential client folders, consultants, and small teams that need shared clarity — without enterprise complexity or cloud AI fees for everyday inference.

For scoping, deployment options, and a confidential discussion of your use case, contact us directly.

contact@realight.dev

Get going

Setup tutorial — click by click

Follow this guide exactly. Button and menu names match what you see in AgentA. Prepare Ollama and models first, then install AgentA. Do the Day-one path for files + Chat, then open each step’s Show setup details for email or Telegram.

Day-one path (about 15–20 minutes)
  1. Install Ollama from ollama.com and leave it running.
  2. Pull the recommended models (commands in step 1 details below) and wait until downloads finish.
  3. Install AgentA from the Microsoft Store → open it → wait for first launch → click Get started.
  4. In AgentA Settings → set Ollama URL + model → Save Settings.
  5. Still in Settings: ➕ Add watch folder… → keep rename/move off → Save Settings.
  6. Open Files → click 🔍 Scan & Index (all watch folders) → wait until it finishes.
  7. Optional: open Document Library and check OCR quality scores on scanned PDFs.
  8. Open Chat → type rag status → set a folder Scope if you like → then ask about one of your files.
  9. Later: try Editor mode — instruct AgentA, review the reworked text, confirm, and save a new .docx — or open a second chat session for another topic.

Continuous watching (laptops): If you want AgentA to keep watching files, email, or Telegram for long periods, prevent Windows sleep with Microsoft PowerToys → Awake (details in the step below). Sleep or aggressive power saving can pause background work.

  1. Install Ollama and pull models Do this before AgentA — local AI must be running with models ready.
    Show setup details

    Goal: Ollama is installed, running in the system tray, and the recommended models are fully downloaded.

    1. Install Ollama and start it

    1. On your PC, open a browser and go to ollama.com.
    2. Download and install Ollama for Windows.
    3. Start Ollama and leave it running in the background (system tray). Keep it running while you use AgentA.

    2. Pull / install the language models — open PowerShell or Command Prompt and run these commands one by one:

    Main model (recommended — chat, vision, Telegram audio; faster):

    ollama pull hf.co/unsloth/gemma-4-e4b-it-GGUF:Q4_K_M

    Optional larger model (same jobs, slower):

    ollama pull hf.co/unsloth/gemma-4-12b-it-GGUF:Q4_K_M

    Embeddings model (needed for smooth private document search / RAG):

    ollama pull nomic-embed-text

    Wait until each command finishes (downloads can take several minutes). Only then continue to install AgentA.

    Why this order: AgentA talks to Ollama on your PC. If you open AgentA before models are ready, Chat will show “Local AI not ready” until you finish this step and connect in Settings.
  2. Install AgentA Microsoft Store install, first launch, trial welcome.
    Show setup details

    Goal: AgentA is open, backend is running, trial welcome is dismissed.

    What you need

    • Windows 10 or Windows 11, 64-bit
    • Ordinary laptop or desktop (no dedicated GPU required)
    • 8 GB RAM minimum · 16 GB recommended · 32 GB best
    • System memory type: DDR4 minimum · DDR5 or newer recommended (dual-channel preferred)
    • Ollama already installed, running, with models pulled (previous step)

    High-end hardware (optional): A dedicated GPU is not required, but when available it can accelerate local inference dramatically. As an example of top-tier equipment — not a baseline requirement — a GPU with about 24 GB of VRAM can run Google Gemma 4 12B at Q8 (8-bit quantization), typically occupying roughly 12 GB of VRAM, with a smooth context window of 32k tokens or more and generation speeds on the order of 50+ tokens per second. That illustrates how responsive AgentA can feel on high-end hardware while remaining fully local.

    Do this

    1. Open the Microsoft Store app.
    2. Search for AgentA (publisher: Realight Dev).
    3. Click Get / Install, then Open.
    4. If Windows Firewall asks, allow AgentA on a Private network.
    5. Wait while you see AgentA is starting... — first launch usually takes 1–2 minutes (backend unpack). Do not close the app.
    6. When Welcome to AgentA Personal appears, read the trial note and click Get started.

    Where things are in the left menu

    • Dashboard — status overview
    • Chat — Local RAG, scoped asks, live web
    • Emails & Messages — priority mail · Email RAG
    • Files — Scan & Index · RAG search index
    • Document Library — OCR quality scores · Re-index RAG
    • Settings — models, folders, Domain & RAG, email, Telegram, plan
    • Also: Urgent Actions, Daily Leads, Registers, Activity, Logs
    • Deals (Deal Map) is Business-only — not part of Personal Store plans

    You start a 14-day Personal Pro trial. Check or upgrade later in Settings → Plan & billing.

    If stuck: Close all AgentA windows. Wait 10 seconds. Open AgentA again. If it still never leaves “starting”, open Logs in the left menu, or contact contact@realight.dev.
  3. Keep the PC awake for continuous watching (recommended) Especially on laptops — stop Windows sleep from pausing AgentA’s file, email, and Telegram watchers.
    Show setup details

    Goal: Windows does not put the PC to sleep while you want AgentA to keep watching and operating.

    Why this matters

    AgentA runs on your PC as a desktop app with local watchers (folders, optional email, optional Telegram). On many laptops, Windows power saving can put the machine to sleep or throttle background activity after idle time. When that happens, AgentA can stop reacting to new files or messages until you wake the PC again. This is normal OS battery behavior — not an AgentA crash.

    If you need AgentA available for long stretches (including overnight “always on” monitoring), keep the PC awake while AgentA is in use.

    Recommended: Microsoft PowerToys → Awake

    PowerToys Awake keeps the computer from sleeping without permanently changing your Windows power plan. Official overview: PowerToys Awake (Microsoft Learn).

    1. Install Microsoft PowerToys from the Microsoft Store (or the official PowerToys release).
    2. Open PowerToys Settings → select Awake.
    3. Turn Enable Awake ON.
    4. Choose a mode that fits your use:
      • Keep awake indefinitely — best for continuous AgentA watching while the PC is plugged in
      • Keep awake for a time interval — for a long scan, download, or work session, then return to normal sleep
    5. Optional: use the coffee-cup Awake icon in the Windows system tray to switch modes quickly.

    Practical tips

    • Prefer plugged-in power for true long-running / overnight use — Awake still consumes power.
    • You do not need to keep the screen bright forever; Awake can keep the system awake while you choose whether the display stays on.
    • Closing the laptop lid may still sleep some devices depending on Windows lid settings — for desk use, leave the lid open or set “When I close the lid” to Do nothing on plugged-in power (Windows power options).
    • Also keep Ollama running in the background whenever you want Chat and local AI available.
    Summary: For reliable 24/7-style watching on a laptop, install PowerToys early, enable Awake, and run AgentA (and Ollama) while the PC stays awake — otherwise Windows sleep can quietly pause file and message monitoring.
  4. Connect AgentA to Ollama In Settings, select your model and Save — required before Chat works.
    Show setup details

    Goal: Settings shows your Ollama URL and an installed chat model, and you clicked Save Settings.

    Ollama should already be running with models installed (step 1).

    1. In AgentA, open the left menu → Settings.
    2. Find the card LLM Configuration.
    3. Set Ollama URL: to exactly http://127.0.0.1:11434 (this is the default).
    4. Open the Model: dropdown and select the E4B model you pulled (full name from the list). If the list is empty, wait a few seconds, or type the model name manually.
    5. Scroll to Images (vision LLM):
      • Leave Use vision LLM for images checked (ON) — default.
      • Leave Optional vision model empty to reuse the main chat model.
      • Leave timeouts at defaults unless images fail later: Vision timeout 300 sec (range 30–3600), Max image edge 2048 px (512–8192), Max tokens 4096 (256–32768).
    6. Optional later — URL summary (for long web/Telegram links): URL timeout default 420 sec (60–3600); URL input max chars default 12000 (2000–60000). Raise timeout if a 12B model times out.
    7. Click the button at the bottom: Save Settings.

    PDF text uses built-in Tesseract OCR (EN, DE, RO, UK). Photos and screenshots use Vision LLM.

    If you see “Local AI not ready” / “Select your local AI model”: click Settings on the banner, confirm Ollama is running, pick a model from the dropdown, click Save Settings. If Save warns the model is not installed, go back to step 1 and pull the model again.
  5. Add a watch folder (keep files safe) Tell AgentA which folder to read — without renaming or moving anything yet.
    Show setup details

    Goal: at least one watch folder is listed, rename/move stays off, Save Settings succeeds.

    Settings → File Watching

    1. Stay on Settings.
    2. In File Watching, click ➕ Add watch folder….
    3. Pick a folder you own (for example a project folder or Documents subfolder).
    4. Put only 1–5 PDFs or Office files there for the first test so indexing is quick.
    5. The first folder becomes primary (★). You can add more folders later and use Set primary / Change / Remove.

    Settings → Automation (important for safety)

    1. Leave Allow AgentA to rename and move files on disk (auto-organize) unchecked (OFF) — this is the default and the safe first choice.
    2. Leave Watch subfolders (new files in subdirectories) unchecked unless you intentionally want subfolders watched (default OFF).
    3. Leave Require confirmation for risky operations checked (ON).
    4. Click Save Settings at the bottom.

    AgentA will index and answer questions in place. Turn rename/move on later only if you want physical reorganization.

    If Save fails: you must have at least one watch folder. Add one with ➕ Add watch folder…, then Save again. You cannot remove the last remaining folder.
  6. Scan & Index, then Chat Build your private index, then ask questions with exact Chat commands.
    Show setup details

    Goal: Scan & Index finishes, then Chat answers a question about your files.

    A. Files page — index safely

    1. Left menu → Files.
    2. Click 🔍 Scan & Index (all watch folders).
    3. Wait until the job finishes. Progress can note native text or OCR character counts and quality. Use ⏹️ Stop only if you need to cancel.
    4. Do not click 📁 Process All in Watch Folder on day one — that path can rename/move files when auto-organize is enabled later.

    Optional: 📂 Scan subfolder… indexes one tree only. Under RAG search index, use Refresh status to confirm chunks exist; avoid Clear RAG index unless you mean to rebuild.

    B. Document Library — OCR quality (recommended)

    1. Left menu → Document Library.
    2. Check tags such as Native text, RAG indexed, and OCR good / fair / weak / poor with a percent score.
    3. If you see ⚠️ OCR review needed, open the file, fix or accept the scan quality, then Mark OCR reviewed or Re-index RAG.
    4. Optionally assign a document type so Chat gates similar files correctly.

    C. Chat page — verify it works

    1. Left menu → Chat (near the top of the sidebar; title: AgentA Assistant).
    2. In the message box, type exactly: rag status → press Enter or Send.
    3. Optional: click 📁 (Scope) → pick a folder or file chip for sharper answers.
    4. Then try: rag: what documents do I have? or search my data: invoice — or a numbered multi-ask (2–4 questions in one message).
    5. Or ask in plain English about a file you just indexed. Expand Sources under the reply to open cited files.

    D. Optional — Revise & Export and multi-chat

    1. Scope one PDF, DOCX, or TXT → instruct AgentA in plain English what to update and export a new version as docx (or pdf/txt).
    2. Review the reworked text on the Before / After card → click Confirm export or type Confirm revise export. You approve; you do not edit the file in Chat.
    3. Find the new sibling next to the original (e.g. … ver.1 DD.MM.YYYY.docx) — the original is never overwritten.
    4. Use the session picker for New / Rename / Delete chats (up to 100 messages each). Leaving Chat keeps it warm for a faster return.

    Useful prefixes (also listed under Chat & Telegram commands)

    • search my data: / rag: — your private files and indexed email
    • web: — live internet (DuckDuckGo-based by default; ephemeral)
    • Paste https://… — fetch and summarize a page
    • offline: / local: — force local-only for that message
    • Confirm revise export — write the versioned file after a revise preview

    Also explore after Chat works

    • Document Library — OCR scores, Re-index RAG, document types
    • Settings → Domain & RAG — Commodity Trading / Generic Business · entity aliases
    • Registers — invoices and contracts
    • Report an issue in Chat — report inappropriate AI content by email
    If Chat says nothing useful: confirm Scan & Index finished, check OCR quality on the file in Document Library, set a folder scope in Chat, confirm Ollama is running, and Settings → Model is saved. Try rag status again. Check Logs if errors persist.
  7. Tune live web search (optional) Choose when AgentA may search the public internet from Chat.
    Show setup details

    Goal: you know whether live web is Automatic, Manual, or Off — and how to force a search.

    Settings → Live web search

    1. Open Settings → card Live web search.
    2. Leave Enable live web search checked (ON) unless you want no web at all.
    3. Set When to search automatically:
      • Automatic (default) — may search when the question looks “live”
      • Confirm / Manual (web:) — only when you type web:
      • Off — no automatic search; you can still type web: to force
    4. Leave SearXNG URL (optional) empty for the default DuckDuckGo-based path.
    5. Click Save Settings.

    You can also change Live web from the Chat toolbar without opening Settings. Check status anytime with web status.

    Try: web: weather in Zurich today · or offline: summarize this from my files only

  8. Connect email (optional) One IMAP mailbox for priority mail, drafts, and urgent tasks.
    Show setup details

    Goal: one email account is created, Test succeeds, watcher is on if you want auto-checks.

    Email is optional. Core Files + Chat work without it. Personal Pro / trial allows 1 email account.

    Prepare an app password first

    • Gmail / Outlook / Microsoft 365: create an app password in your account security settings. Do not use your normal login password.
    • Other providers: use IMAP credentials your host provides.

    Settings → Connected Accounts

    1. Open Settings → scroll to Connected Accounts.
    2. Click + Add Account.
    3. Set Account Type: to Email.
    4. Fill Account Name: (example: Work Email).
    5. Choose Email Provider: Gmail, Outlook, or Custom (IMAP/SMTP).
    6. Enter Email Address: and App Password:.
    7. For Custom: IMAP port default 993 (SSL/TLS), SMTP port default 587 (STARTTLS) — change only if your provider says otherwise.
    8. Set Check Interval (minutes): how often AgentA looks for new mail — default 15 (allowed 1–1440, max 24 hours).
    9. Optional checkboxes:
      • Enable email watcher (auto-check for new emails) — turn ON for automatic checks
      • Automatically download email attachments — default ON for new accounts
      • Auto-reply to WORK emails… — creates drafts only; you still send manually
    10. Click Create Account.
    11. On the new account row, click 🔍 Test. Fix credentials if Test fails.

    Then open left menu Emails & Messages for priority mail. Urgent items also appear under Urgent Actions.

    Common failure: “login failed” almost always means you used a normal password instead of an app password, or IMAP is blocked by the provider. Create a fresh app password and Test again.
  9. Connect Telegram (optional) Create a bot, paste the token, lock it to your user ID.
    Show setup details

    Goal: Telegram bot account is connected, Test works, and only your user ID can run commands.

    Telegram is optional. Personal Pro / trial allows 1 Telegram bot.

    1. Create the bot in Telegram

    1. Open Telegram and message @BotFather.
    2. Send /newbot and follow the prompts (name + username ending in bot).
    3. Copy the bot token BotFather gives you (keep it private).
    4. Start a chat with your new bot and tap Start.

    2. Add the bot in AgentA

    1. Settings → Connected Accounts → + Add Account.
    2. Account Type: Telegram Bot.
    3. Enter an Account Name and paste Bot Token.
    4. Set Check Interval (minutes): default 15 (range 1–1440).
    5. Optionally enable Enable watcher (auto-check for new messages) and auto-reply for WORK messages.
    6. Click Create Account, then 🔍 Test.

    3. Authorize yourself (required for a private bot)

    1. In Telegram, open @userinfobot (or similar) and send /start to see your numeric User ID.
    2. In AgentA Settings → Telegram Authorized Users, paste that ID into Add Telegram User ID:.
    3. Click + Add User.

    If the authorized list is empty, AgentA shows that everyone can use bot commands — add your ID to lock it down.

    Optional: in Automation, leave Enable daily leads capture from Telegram screenshots ON if you want leads from images.

    Same chat commands work in Telegram — see Chat & Telegram commands. Send one message at a time and wait for the reply.

    If the bot ignores you: confirm you pressed Start on the bot, the token is correct, watcher is enabled, and your numeric ID is in Telegram Authorized Users.
  10. Plan & billing See trial days, upgrade Monthly/Lifetime, or restore Store purchases.
    Show setup details

    Goal: you know your current plan and how to upgrade or restore.

    Settings → Plan & billing

    1. Open Settings → card Plan & billing.
    2. Read Current plan and any Trial: N day(s) remaining line.
    3. To upgrade: click Upgrade to Personal Pro → choose Personal Pro Monthly or Lifetime → complete the Microsoft Store purchase dialog (one screen).
    4. If you already paid but Pro is not unlocked: click Restore Microsoft Store purchases.

    Deal Map / team features are AgentA for Business — not Personal Store add-ons.

    Support: contact@realight.dev · Privacy: privacy policy

Day-one safety rule: keep Allow AgentA to rename and move files on disk (auto-organize) unchecked, and use 🔍 Scan & Index (all watch folders) — not Process All — until you deliberately want to rename files or reorganize storage of your files.

Vocabulary

Chat & Telegram commands

Type these in Chat or send them to your AgentA Telegram bot (authorized users only). Same language in both places. AgentA Chat has three modes — Librarian, Analyze (LLM), and Editor — over your indexed files, plus optional live web.

Three Chat modes (overview)

Librarian

Ask any question about indexed document content. Analyze toggle off. Grounded answers + Sources from your private index.

Analyze (LLM)

Scope an indexed file (or compare docs). Ask for text improvement, review, or comparison on parameters you set. Analyze toggle on.

Editor

Instruct AgentA (plain English) to update a found or Scoped document → review the reworked text → confirm save as .docx or PDF. You approve; you do not edit the file in Chat. Full flow: Chat modes.

Your files & mail (local RAG)

search my data: invoice Acme 2024

Fast search in your indexed files, emails, and tasks — ranked chunks, no long AI essay.

rag: what contracts mention delivery dates?

Ask over your private knowledge base. Answers cite real Sources (paths, page hints, OCR badges when text came from OCR).

rag: latest email about payment

Email RAG: grounded answers from indexed mail. In Chat, open Scope → Emails to pick messages, a sender, or all indexed email.

1) Find the seller on the purchase contract 2) Port of discharge on the B/L 3) List SKUs on the packing list

Multi-ask: number 2–4 questions in one message. AgentA answers each in order and merges Sources. Optional shared prefix: For Acme Corp: before the list.

rag status

Show index readiness: enabled flag, embedding model, chunk counts by source (files / emails / tasks), last indexed.

index my emails and files

Build or refresh the private index so Chat can find your content (files + emails + urgent tasks when enabled).

index emails

Index or refresh email only — useful after connecting a mailbox.

rebuild rag index

Rebuild the private index from scratch when results look stale (e.g. after a major Scan & Index format upgrade).

scan and index

Scan your watch folder and index documents in place (non-destructive). Same idea as 🔍 Scan & Index on the Files page.

Get better Local RAG answers

📁 Scope → Use this folder / file

Biggest accuracy win: set a folder or file chip before you ask. Also: Compare these 2 files, or Analyze up to 5 selected files.

Acme Corp contract payment terms

Name the counterparty + document type + clear field (port, quantity, expiry, invoice total, country of origin). Vague “payment terms” alone is weaker.

Analyze with LLM · off

Librarian mode (Analyze off): fast grounded field facts from indexed text. Turn Analyze on for fuller LLM summaries on scoped docs. Analyze is independent of Revise & Export.

Reply 1 or 2

If Chat shows a short list of candidate files, reply with the number — don’t restart with a new long question.

Settings → Domain & RAG

Pick Commodity Trading or Generic Business, add entity aliases, and optional custom field intents so questions match your naming.

Document Library · OCR quality

After Scan & Index, each file can show OCR good / fair / weak / poor + percent. Review weak scans, Mark OCR reviewed, or Re-index RAG so Chat cites cleaner text.

Editor — Revise & Export (.docx / PDF)

📁 Scope → one PDF / DOCX / TXT

Beyond RAG Q&A: pin exactly one file to edit. Folder-only scope is not enough for export. Optional: Analyze the clause first — that does not open a revise card.

Revise this document: rewrite … then export a new version as docx

Editor flow: find or Scope the file → instruct AgentA in plain English what to change → review the proposed reworked text (Before / After) → Confirm → AgentA saves a new .docx or PDF (or txt). You do not edit the document in Chat. Original untouched.

Confirm revise export

Required before anything is written. Or tap Confirm export on the card. Without preview + confirm, AgentA will not create a file.

ContractName ver.1 12.09.2026.docx

New sibling next to the original: {stem} ver.N DD.MM.YYYY.docx (or .pdf / .txt). N auto-increments; date is the export day. Never overwrites the signed/original file.

Multi-chat & session controls

New / Rename / Delete chat

Keep separate titled conversations (e.g. one per client or deal). Each session stores up to 100 messages (oldest trimmed).

Clear chat

Wipe the current session history when you want a clean slate. Per-message Delete and Edit / Save & resend stay available on turns.

Leave Chat → come back

Keep-alive: after the first Chat visit, leaving the section hides Chat instead of tearing it down — the session list returns instantly.

Live web & URL summarize

web: weather in Zurich today

Web RAG: searches the live internet with DuckDuckGo, reads useful pages, then your local LLM answers with source links. Fetches are ephemeral — not added to your private RAG index by default.

search the web: …

Same as web: — alternative wording.

https://example.com/article

Paste a URL — AgentA fetches and summarizes the page locally (same idea as analyze URL). Pair with local RAG when you want “my files vs this article.”

offline: summarize this from my files only

Keep the answer on local RAG only — skip automatic web search for this message. local: works the same way.

web status

Check web-search settings (also useful from Telegram). Policies: Automatic · Manual (web:) · Off.

Telegram extras

rag what invoices are overdue?

On Telegram you can omit the colon — rag … is treated like rag:.

Send a PDF / photo / link

AgentA returns a structured summary (TLDR, key points, details, caveats).

Voice note

Local speech-to-text → actionable urgent tasks (with a vision/audio-capable model).

extract text

As a caption on a file: dump OCR/raw text only (no field extraction).

One message at a time

Wait for the reply before sending the next request — local AI works best that way. One AgentA instance should poll each bot token.

Modes in one Chat: Librarian = ask indexed content · Analyze (LLM) = scoped review, improvement, or compare · Editor = instruct AgentA → review reworked text → confirm new .docx/PDF · web: / paste URL = live public lookup. See Three modes in AgentA Chat. Use 📁 Scope and clear wording for precision; keep the index healthy with Document Library OCR scores. Multi-chat keeps client or deal threads separate. Plain English also works — prefixes, Scope chips, and Confirm keep intent unmistakable.

FAQ

Common questions

Quick answers about AgentA Personal, privacy, pricing, and setup.

What is AgentA?

AgentA is a private AI employee for Windows. It indexes your files for versatile Chat RAG — ask your documents, edit text, and save a new .docx (or PDF/TXT) — while you keep your own file storage. Rename/move on disk is optional and off by default. Optional email triage, tasks, and live web. Your data stays on your PC.

Does AgentA send my files to the cloud?

AgentA is local-first. Documents, chat history, and indexes are stored on your Windows PC. Inference runs through Ollama on your machine unless you configure a remote endpoint yourself.

What do I need to run AgentA?

Windows 10 or 11 (64-bit), AgentA from the Microsoft Store, and Ollama with a local model such as Google Gemma 4. Memory guidance: 8 GB RAM minimum, 16 GB recommended, 32 GB best; DDR4 minimum, DDR5 or newer recommended. A dedicated GPU is optional — high-end hardware (for example about 24 GB VRAM) can substantially accelerate local inference. See the Setup tutorial for the full checklist and a top-tier performance example.

How much does AgentA cost?

AgentA Personal is free to install with a 14-day Personal Pro trial. Optional Microsoft Store add-ons include Personal Pro monthly and lifetime plans.

What is local RAG?

RAG (retrieval-augmented generation) means AgentA searches your indexed files and emails first, then answers from those results — not the open web by default. Chat supports folder/file/email Scope, numbered multi-ask, foldable Sources (with page hints and OCR badges), Librarian vs Analyze modes, streaming controls with live tok/s, multi-chat sessions, and Revise & Export for versioned PDF/DOCX/TXT. After Scan & Index, Document Library shows an OCR quality score (good / fair / weak / poor + percent) so you can review noisy scans and re-index when needed. See Chat & Telegram commands for prefixes and tips.

What are the three Chat modes?

Librarian — ask any question about the content of your indexed documents (Analyze off).
Analyze (LLM) — Scope an indexed file and request deeper analysis: text improvement, review, or compare documents against parameters you specify (Analyze on).
Editor — instruct AgentA in plain English to update a found or Scoped document; review the reworked text it proposes; confirm to save a new .docx or PDF. You do not edit the document in Chat — original untouched.
Full detail: Three modes in AgentA Chat.

What is Revise & Export?

Editor mode: find or Scope the document → instruct AgentA in plain English what to update → review the proposed reworked text (Before / After) → confirm with the button or Confirm revise export. AgentA saves a new version as .docx or PDF (TXT too), e.g. ContractName ver.1 12.09.2026.docx. You approve the rewrite; you do not edit the file in Chat. The original is never overwritten. Normal Q&A and Analyze do not open a revise card.

Can I keep several Chat conversations?

Yes. Use New / Rename / Delete in the Chat session picker. Each titled chat holds up to 100 messages. After you open Chat once, leaving for Files or Settings keeps Chat warm (hidden, not destroyed) so returning is instant.

What is the OCR quality score?

During indexing, AgentA prefers a native PDF text layer when present; otherwise it runs Tesseract OCR (with optional Vision LLM for photos). Document Library labels each OCR result as good, fair, weak, or poor with a percent score. Weak or poor files can be reviewed, marked reviewed, or re-indexed so Local RAG Chat cites cleaner text.

Is AgentA available on macOS?

AgentA for Windows is available on the Microsoft Store today. AgentA Max for Apple Silicon (local MLX, unlimited My Agents) is in development — see the AgentA Max section and the blog post.

Where can I download AgentA?

Get AgentA from the Microsoft Store for Windows 10 and 11.

Ready for a private AI employee?

Install from the Microsoft Store. Index your files for versatile Chat RAG — you keep your own storage. Ask, edit, save .docx. Get your time back.

Get it from Microsoft Setup guide
Benefits Get it from Microsoft