Screen observation
Use image matching, OCR, object detection, screenshots, and UI tree search to find text, buttons, templates, or visible targets on the current Android screen.
vision.matchvision.ocrvision.detectvision.captureui.findLaiCai Flow: AI Android Automation Tool
LaiCai Flow helps you use AI to automate Android devices and emulators. You can edit Flows in Graph View, generate Flows by chatting inside the LaiCai app through an LLM API, or let Codex, Claude, and other desktop AI tools generate Flows through MCP.
More LaiCai Flow video tutorials
- LaiCai Flow UI overview
- Create your first LaiCai Flow profile
- Create your first LaiCai Flow node
- Build a LaiCai Flow manually
- Generate a LaiCai Flow with AI Planner
- Generate LaiCai Flow profiles with Codex and MCP
- Run Android automation on devices with LaiCai Flow
- Automate Google Photos with LaiCai Flow nodes
- Delete one Google Photos image with LaiCai Flow nodes
- Delete multiple Google Photos images with Flow repeat nodes
- Delete all Google Photos automatically with flow.until
- Auto type text on Android with data.array and flow.foreach
Core concepts
Think of a Flow configuration as a flowchart: each node in the Flow does one thing. Every node has inputs and outputs, and the previous node passes data to the following nodes. For example, an [Image Match node] only finds the image coordinates, then passes those coordinates to the next [Click node] to perform the click task.
Profile and Main Flow
A profile stores one or more flows. The Main Flow is the entry point that runs when you start automation on a device.
Nodes and data
Flow provides many types of nodes for different kinds of tasks. Each node handles one clear action or data-processing step, such as screen operations, conditions, data transformation, flow control, external service interaction, or calling another Flow. After a node runs, it can output data for later nodes to use.
What automation nodes can do
LaiCai Flow nodes cover screen observation, device actions, data processing, flow control, and external integrations. You combine them like building blocks: observe the screen, choose data, act on the device, then branch, wait, loop, or call another Flow.
Touch and input
Tap, swipe, long press, drag, type text, press keys, or use the clipboard after a previous node provides a target position or value.
pointer.tappointer.swipepointer.longPresspointer.draginput.textApp and device control
Open or stop Android apps, send Home and Back, take screenshots, record the screen, adjust volume, or save media to the phone.
app.openapp.forceStopdevice.homedevice.backdevice.screenshotData processing
Pick fields, read JSON paths, build objects and arrays, filter and sort lists, compare values, and store runtime state for later steps.
data.pickdata.filterdata.sortdata.jsonGetvar.setstorage.getFlow control
Branch by conditions, wait between steps, stop a run, call child Flows, repeat a Flow, loop over arrays, or run until a child Flow succeeds.
flow.branchflow.waitflow.runflow.repeatflow.forEachflow.untilExternal integration
Call HTTP APIs, load local files, run saved macros or AutoJS scripts, and use ADB Shell when a task needs deeper Android automation.
http.requestfile.loadmacro.playautojs.runadb.shellThree ways to create a LaiCai Flow automation workflow
Method 1
Edit a Flow in Graph View
Best for people with programming or automation experience when the task is clear and not too complex. In Graph View, manually add automation nodes, connect them in execution order, then configure each node’s parameters, inputs, outputs, and data passing between nodes.
- Break the goal into a few deterministic steps, such as open app, wait for screen, and tap a button.
- Use connections to show the execution path and simple branches, such as success, failure, true, and false.
- Run after each small section and check the log for node parameters, inputs, outputs, and device state.
Method 2
Generate a Flow by chatting inside the LaiCai app
No programming background is required. Anyone can describe an automation task in natural language and let AI generate or edit a Flow inside the LaiCai app. You can explain the goal like a conversation, such as opening an app, finding a button, reading text, tapping an element, entering content, or repeating actions. You can also use @mention to quickly choose template images, OCR regions, device assets, or an existing node so AI can update the current Profile more accurately.
- Describe the task in plain language, such as “open YouTube, tap the search icon, and enter LaiCai Flow”.
- Use @mention to choose assets such as template images, recognition regions, button screenshots, or existing nodes.
- AI uses the current Flow, node capabilities, device state, and selected assets to generate or adjust node connections and parameters.
- If package names, template images, OCR regions, account details, or API parameters are missing, the app asks for them instead of guessing.
Method 3
Generate Flows through MCP in Codex, Claude, and other desktop AI tools
Best for people who already use Codex, Claude, or similar desktop AI tools. Through LaiCai MCP, the desktop AI can directly read devices, node capabilities, automation assets, and existing Profiles, so you do not need to copy that context into the chat and do not spend extra LLM API tokens on that local context. You can keep using Codex and Claude for code reading, file analysis, structured reasoning, and long-context work while AI generates, checks, or updates a LaiCai Flow Profile.
- No need to manually copy device information, node schema, assets, or Profile content into the chat window.
- No extra LLM API tokens are spent on local context provided through MCP.
- Combine Codex and Claude code analysis, file reading, and problem breakdown to design more complex Flows.
- Useful when AI should reference local project files, automation assets, and real device state while maintaining a Profile.