CORE FEATURES

Core Features

From basic automation to intelligent workflows, ACRPA covers the entire desktop automation stack.

📦

Excel-driven scripts

Uses .xls spreadsheets as script files — commands in column A, parameters in columns B–J. Orchestrate automation without programming; the learning cost is minimal.

🖱

Image-based localization

Image recognition built on OpenCV / Pillow: full-screen and region-limited search, grayscale accelerated matching — locate and click any UI element precisely.

🪟

Window management

Operate Windows directly: activate, close, minimize, wait for a window to appear or disappear. Window-relative coordinates make it 10–100× faster than image matching.

🎬

Record & replay

Record mouse and keyboard actions (including drag, scroll and window activation), with wait intervals inserted automatically. Replay in one click — repetitive work is never manual again.

⏱

Scheduled tasks

Daily / weekly / interval / one-shot scheduling modes, executed silently in the background alongside the system tray and Mini Bar.

⚡

DD driver boost

Optional kernel-level high-performance input backend that simulates real keyboard and mouse input and is hard to detect. Supports direct / simulate / auto write modes with automatic fallback.

🧩

Workflow orchestration

Chain multiple scripts with a visual flow diagram (nodes / connectors / drag-and-drop) supporting four step types: script, parallel, condition and wait.

📊

Variable system

Set variables, read the clipboard, substring / replace, math operations — ${var} syntax works throughout, so complex data is easy to handle.

🔍

OCR recognition

Recognize text in a screen region, wait for text to appear, click on text. Compatible with PaddleOCR / Windows OCR / Tesseract — even dynamic interfaces can be automated.

🐞

Professional debugger

Breakpoints / conditional breakpoints, single-step, variable watcher, call stack, execution timing, run-to-cursor and skip-line — problems become obvious.

🌓

Dark mode

Switch between light and dark themes in one click. With the system tray and the Mini Bar floating status strip, automation runs quietly in the background.

📜

Logs & marketplace

Structured levelled logs with daily rotation, automatic cleanup and one-click export; built-in script marketplace (works offline) that keeps expanding the script ecosystem.

🔄

Automatic updates

Self-checks for new versions at startup and fetches the manifest with multi-source fallback (GitHub Releases → raw manifest → CDN mirror). Downloads show progress and are verified by sha256; if a mirror serves an old package it switches source automatically, then a one-click restart completes the replacement without touching your config or scripts.

🌐

Browser automation

Built-in Playwright web automation: open a page, click, type, wait for elements, full-page or element screenshots — desktop and web steps can be chained in the same script.

🛡️

Safety guards

Runaway protection (flick the mouse to abort immediately), maximum execution time limits, stop-on-error and failure retry policies — unattended runs still end safely.

NETLINK · MULTI-DEVICE

One PC, Many LAN Machines · NetLink

See the progress of every machine on the LAN at a glance, run / pause / stop remotely, distribute scripts in bulk, and view it all from a phone browser — built entirely on the Python standard library with zero extra dependencies, a multi-machine control plane rarely found in tools this lightweight.

⚡ Zero extra dependencies (pure stdlib) 📱 Phone browser dashboard 📦 Bulk script push · run after push 🔐 Pairing code + privilege levels + audit log

🖥️

Real-time multi-machine monitoring

Device list plus run state, current script, progress (line number / loop count / elapsed time), live log stream and scheduled task status — select a device and see it immediately, no manual subscription needed.

🔐

Pairing authentication

Enter a 6-digit pairing code (valid for 600s by default) to pair. The PIN never travels over the network: a PBKDF2-HMAC-SHA256 derived key plus nonce/HMAC challenge-response is used, and 5 consecutive failures lock the connection. Keys are stored in the Windows credential store, so reconnects after pairing need no PIN.

🎛️

Remote control

Run / pause / resume / stop. The controlled machine prompts for confirmation on first control (this device can be remembered). Commands are serialized and acknowledged to all controllers; every remote command and screenshot request is written to the audit log.

📦

Script distribution

Push local .xls / .xlsx / .json files to controlled machines: 64 KB chunking with dual sha256 and size verification on the final frame, leaving no leftover files on failure. Supports bulk delivery to multiple devices and "run right after push".

📱

Read-only browser panel

A built-in single-page panel served by http.server (default port 19712, off by default). Check multi-machine progress and logs straight from a phone or tablet browser; token auth (query string / cookie / bearer) with zero write operations.

🛡️

Optional TLS encryption

Encrypted channel with a self-signed certificate based on the stdlib ssl module; the peer certificate sha256 fingerprint is pinned (TOFU) and mismatches are rejected outright. Off by default, enable as needed.

Multi-device · device list and live progress

NL 01 · Device Linking

All machines, one screen

Device list, run state, current script, line number / loop count / elapsed time and the live log stream all in one window — see the whole picture without switching machines.

Multi-device · pairing and first-control confirmation

NL 02 · Pairing

PIN + challenge-response + credential store

The controlled machine generates a 6-digit pairing code and the controller enters it to pair. Keys are never written to config.json, reconnects need no PIN after pairing, and first-time control still requires confirmation on the controlled side.

NL 03 · Browser Panel

Viewable from a phone browser

A built-in read-only web panel (default port 19712, off by default). Scan a QR code or type the address on a phone or tablet to view multi-machine progress and logs — the panel contains no write operations at all.

COMMAND SYSTEM

Command System · 51 Built-in Commands

51 built-in commands: write one per row in Excel and use them right away. Click any command or example to copy it. See the README → for full documentation.

Expand the Command Reference · 7 categories · 51 built-in commandsFull docs in the README →

🎯 Basic Ops

🪟 Windows

🧮 Variables & Logic

🔍 OCR

🤖 AI

🧩 Workflow

🌐 Browser

Command	Description	Example
`找图`	Search the whole screen for an image; on match, move the mouse to that position (hover)	找图, hello, 0.9
`点图`	Find an image and click it, with precision control	点图, submit, 0.9
`区域找图 / 区域点图`	Limit the scan region for faster matching, with grayscale acceleration	区域找图, icon, 0.8, 100,100,500,500
`坐标`	Click at the given screen coordinates; button, click count and interval are configurable	坐标, 500, 300, left, 1, 0.1
`悬停`	Move the mouse to the given coordinates	悬停, 500, 300
`按键 / 热键`	Simulate key presses and combinations (Ctrl+C / Ctrl+V and so on)	热键, ctrl, c
`输入 / 写入`	Paste-style input (unaffected by the IME) or character-by-character simulated typing (DD driver supported)	写入, Hello, 0.05, auto
`等待`	Delay / wait, supports random ranges (for example 1.0-3.0)	等待, 1.0-3.0
`滚轮 / 拖拽 / 相移`	Scroll, drag to given coordinates, or move relative to the current mouse position	滚轮, -3
`按下 / 释放`	Hold and release a key to implement press-and-drag and other complex actions	按下, a
`复制 / 粘贴`	Built-in shortcut combinations (Ctrl+A+C / Ctrl+A+V)	复制
`代码`	Execute Python code from a specified txt file (the program built-in libraries can be called)	代码, my_code
`截屏`	Take and save a screenshot; a timestamp is appended to the filename automatically	截屏, shot, E:/ACRPA/screenshots

Command	Description	Example
`激活窗口`	Activate the window with the given title; fuzzy matching supported	激活窗口, Notepad
`关闭窗口`	Close the specified window	关闭窗口, Untitled - Notepad
`最小化 / 最大化窗口`	Minimize the window to the taskbar / maximize it	最小化窗口, Calculator
`获取窗口位置`	Get the window coordinates and size and save them to variables	获取窗口位置, Notepad, x, y, w, h
`等待窗口`	Wait for a window to appear or disappear, with a configurable timeout	等待窗口, Chrome, 10, exists
`窗口坐标`	Click relative to the window top-left corner; stays accurate even after the window moves ★	窗口坐标, Notepad, 100, 50

Command	Description	Example
`设置变量`	Set a variable value; numbers are converted automatically and ${var} references work throughout	设置变量, count, 100
`读取剪贴板`	Read clipboard content into a variable	读取剪贴板, clip_text
`字符串处理`	Substring / replace operations on strings	字符串处理, text, substring, 0:5, result
`数学运算`	Perform math calculations; ${var} references are supported	数学运算, ${x}*${y}+1, sum
`如果 / 否则 / 结束如果`	Conditional branch control flow	如果, ${count}>10
`循环开始 / 循环结束`	Loop block, by iteration count or condition expression	循环开始, 5
`跳出循环`	Break out of the current loop immediately	跳出循环

Command	Description	Example
`识别文字`	OCR the text in a given screen region and store it in a variable	识别文字, 0, 0, 800, 300, ocr_text
`等待文字`	Wait for specified text to appear / disappear, with a configurable search region	等待文字, Submit succeeded, 10, exists
`点击文字`	Find the text position and click it; confidence and search region are configurable	点击文字, Login now, 0.8

Command	Description	Example
`AI找图`	Describe a UI element in natural language; AI analyzes the screenshot and locates it (click / hover / find only)	AI找图, the login button, 0.8, click
`AI识别界面`	AI analyzes all UI elements on the current screen; results are stored in a ${prefix}_elements variable	AI识别界面, ui
`AI优化建议`	AI analyzes historical execution data and gives script optimization suggestions	AI优化建议, log
`[AI] 调试`	Ask AI in natural language; AI analyzes the logs and answers debugging questions	click the [AI] Debug button

Command	Description	Example
`运行工作流`	Invoke a workflow file (.json / .yaml) to run multi-script orchestration	运行工作流, workflow.json
`工作流变量`	Set workflow-level shared variables for use by multiple sub-steps	工作流变量, var_name, value

Command	Description	Example
`打开网页`	Open a URL in the built-in browser; suitable for post-login data retrieval, web form filling and similar flows	打开网页, https://example.com
`浏览器点击`	Click a web element by CSS selector or visible text	浏览器点击, #login-btn
`浏览器输入`	Fill text into the specified input box (located by selector)	浏览器输入, #username, admin
`等待元素`	Wait for a web element to appear / disappear, with a configurable timeout	等待元素, .result, 10, 出现
`浏览器截图`	Save a full-page or specific-element screenshot	浏览器截图, result, page

Browser commands are implemented on Playwright, installed on demand (about 155 MB, not bundled with the main program); if it is missing, the command shows how to install it. Desktop and web steps can be chained serially in the same script.

TECH STACK

Tech Stack & Runtime

Lightweight, fast cold start, small footprint. Available as an install-free exe or run from source.

🧰 Tech Stack

Python 3.7+ tkinter PyAutoGUI OpenCV Pillow pywin32 DD driver PaddleOCR Windows OCR Tesseract AI API Playwright

Packaged as a single-file exe — double-click and go. If OpenCV is unavailable it falls back to Pillow matching automatically, so it stays stable instead of crashing.

Full documentation / README →

⚙️ Key Configuration (config.json)

dark_mode Light / dark theme switch

retry_max / retry_interval Failure retry count and interval

image_timeout Image match timeout in seconds

api_key / api_model AI key and model selection

use_dd_driver DD driver high-performance input

minimize_to_tray / mini_bar_enabled Tray residency and collapsible status bar

sched_enabled / sched_tasks Scheduled task execution

log_level / log_retention_days Log level and retention days

stop_on_error / failsafe Stop-on-error and runaway protection

check_update / app_protect Startup update self-check and process guard

CHANGELOG

Release History

Continuous iteration, recently focused on the update mechanism, workflow orchestration, AI enhancement and stability. See GitHub Releases → for the full history.

Rebuilt on the v0.1.25 baseline to fix the unstable wiring introduced in v0.1.26: an undefined _update_pending made _periodic throw a NameError every 100 ms, which broke the status-bar and execution-progress refresh path and prevented the update flag from being set; and an unimported show_update_dialog made the status-bar update entry point dead on click. All v0.1.26 improvements are retained: the reworked update mechanism (multi-source fallback / semantic version comparison / sha256 verification / one-click restart self-update for the portable build / fixed version source for the packaged build), engine-level command failure semantics with workflow error propagation, and the release toolchain with build assets under management.

v0.1.26

2026-09-13

Update mechanism reworked: fixed the distorted version number after packaging (the EXE always reported an old version, breaking update checks and the About page entirely). Update checks now use multi-source fallback (GitHub Releases → raw manifest → CDN mirror) and return as soon as one source works; if a source reports a version that is not new enough, the remaining sources are probed and the highest is taken, so a stale mirror cache or a network glitch no longer yields a false "already up to date". Version comparison handles the v prefix, uneven segment counts and pre-release suffixes. Network transport is dual-stack (requests → system certificate store), fixing silent failures behind corporate TLS interception. Downloads retry multiple candidate direct links one by one, stream progress, verify sha256 and size, and re-check the VERSION inside the package, switching source automatically when a mirror serves an old package. A new in-app update window and one-click restart self-update for the portable build replace only the main program and VERSION — config, scripts, templates and logs are never touched. Also fixed "command failure silently reported as success" together with the broken stop_on_error and retry behaviour.

v0.1.25

2026-08-10

Deep customization of the Workflow tab: categorized action tree with search and recent items; new command / variable / loop / log nodes sharing one configuration form, with right-click copy / paste / disable / comment; control over outer loop count and maximum execution time; execution highlighting and variable management; centrally managed version numbers.

v0.1.24

2026-08-09

New Advanced Settings page (maximum execution time / globally bound window / recording mode / OCR backend / headless browser and more); every settings card gets its own "Apply ✓"; the model dropdown now uses a unified registry (DeepSeek / Qwen / GPT / Claude); fixed the DD DLL path not taking effect and the wrong Claude endpoint mapping.

v0.1.23

2026-07-20

Template picker dialog switched to a top-bottom layout; exception protection when toggling the tray; a "restore defaults" button; fixed issues such as the script not loading after recording and the card_log variable name collision.

v0.1.22

2026-07-20

Structured logging upgrade (log levels / timestamps / daily rotation / automatic cleanup); settings page rebuilt as six independent cards; a record button added to the script editing toolbar.

v0.1.21

2026-07-18

System tray icon with balloon notifications and a context menu; Mini Bar collapsed mode (a floating always-on-top status strip) that is draggable, follows the theme and updates its status live.

v0.1.20

2026-07-17

Visual workflow diagram (nodes + connectors + drag and drop); window-relative coordinate system; recording enhancements (drag / scroll / window activation); improved conditional breakpoint UI and LRU eviction for the image cache.

v0.1.19

2026-07-16

AI enhancement module released: visual localization (AI find image / AI inspect UI), smart retry, anomaly detection, flow optimization suggestions and natural-language debugging.

v0.1.18

2026-07-06

OCR text recognition commands, a script marketplace panel, debugger enhancements (execution timing / skip line / run to cursor / in-place variable editing) and package size optimization.

v0.1.17 → v0.1.13

2026-06 — 2026-07

Integrated the DD driver high-performance input backend; unified shutdown cleanup; AI generator and template improvements; more complete documentation for image/color recognition and OCR.

FAQ

Frequently Asked Questions

A few things you may want to know before downloading.

Do I need Python installed?

No. Download the portable exe and double-click to run — no Python installation required. You only need Python 3.7+ if you want to modify the source.

Which operating systems are supported?

Windows (x64) desktop only, available as a portable build and an installer. macOS and Linux are not supported at this time.

What format are scripts in?

Excel .xls spreadsheets (.xlsx is not supported yet): one command per row, with the command name in column A and parameters in the following columns. Ten templates and a recorder are built in to generate scripts automatically. Image assets must be PNG, and avoid non-ASCII characters in paths and filenames.

Is it really free? Can I use it commercially?

Completely free under the MIT license. Individuals and companies alike may use, modify and commercialize it. No feature or time limits, no ads, no in-app purchases.

Antivirus or SmartScreen flags it. What now?

Because the binary is not signed with a paid certificate, the first run may be blocked. Click "More info → Run anyway" or add the folder to your allowlist; you are also welcome to build from source and verify it yourself.

Do the AI features cost extra?

ACRPA itself is free. AI uses your own model key (DeepSeek / Qwen / OpenAI / Anthropic / local Ollama and so on) and is billed by that provider based on usage. Without AI it runs fully offline at zero cost.

How is it different from other RPA tools?

It is lighter (single file, no installation), more approachable (an Excel spreadsheet is the script, so anyone can read and write it), supports local AI visual localization, and is fully open source so you can modify the code for free.

Are my scripts and data safe?

Automation runs locally, keys are kept in the system credential store, and the program neither collects nor uploads your operational data.

Who Uses ACRPA

🗂 Office Clerks / Admin

💰 Finance / Bank Reconciliation

📊 Daily Reports & Scheduled Monitoring

🖥 Intranet Multi-Machine Ops

🧑‍💻 Developers / Power Users

🛒 E-commerce · Bulk Listing

Up and Running in 3 Steps

Record it, or write it in Excel

Tune parameters, let AI locate

Run it, then let it run

Interface Overview

Your spreadsheet is the script

Run control and debugging hub

Chain multiple scripts together

Settings cards by category

Light / dark themes, one click