Use dialogue to complete the entire Flows: the agent independently creates nodes according to your needs, writes prompt words, connects and runs the workflow. It can be used directly within Flows, or remotely controlled through MCP from external agents such as Claude, ChatGPT, and Codex.
1.Agents within Flows
Purpose: Use dialogue to describe the goal, and the agent creates nodes in the current Flows, writes prompt words, connects and runs.
Open the agent and issue requirements
Open the agent panel on the right side of Flows and describe the desired result in one sentence; attach reference materials or specify nodes on the canvas if necessary.
01
Open to the right of Flows Flows Agent panel. Displayed when the panel is empty quick start Example, click to apply directly.
Describe the target to the right of Flows
01Session consumption: Displays the points deducted for this conversation, and the fees generated by nodes are calculated separately.
02Requirement input box: Use one sentence to describe the desired result; enter @ to specify a node on the canvas.
03Model, running mode and attachments: Select the model used by the agent, whether confirmation is required, and add reference materials or skills.
02
Describe the goal in the input box. input @ You can specify a node on the canvas; click the paper clip to select from Assets, or upload pictures or videos for reference.
03
If the node is already selected on the canvas, the panel will display Reference Flows selection, indicating that this message will bring these nodes. Click when not needed Ignore Flows selection Ignore this selection.
The execution process of the agent
The agent will first give a plan and then execute it step by step. Each step can be seen in the panel, and the results appear simultaneously on the canvas.
01
Panel display after sending Executing plan Think about the steps and then start executing them, with the number of completed steps displayed in real time at the top.
View execution steps and pending operations
01Current actions and progress: Shows the actions being performed and the number of steps completed.
02Your instructions: Each instruction corresponds to its execution process.
03Waiting for processing: Actions that require your confirmation will stop here and will not be executed on their own.
04Approve or reject: If approved, continue execution, if rejected, skip this step.
05Session consumption: The real-time cost of agent dialogue, excluding node generation costs.
02
Common actions include:Creating node、Updating node、Connecting nodes、Arranging nodes、Starting build、Wait for the build to complete、Locating node。
03
After the generation is completed, the product appears in both the canvas node and the panel. Generate material , you can directly click to view the large picture or video.
Answer questions and confirm changes
The agent will ask questions when the requirements are unclear; modifications to existing nodes or runs that consume points can be set to require your confirmation before execution.
01
Click the run mode below the input box, and Agent permissions Select the confirmation strategy:Automatic execution All done automatically;Confirm before modifying It is only asked before modifying existing nodes and consuming points before running. New nodes and connections are still executed directly.
Choose when the agent needs your approval
01Automatic execution: everything is completed automatically when the balance is sufficient.
02Confirm before modification: Modifying existing nodes and consuming points requires approval first. New nodes and connections will still be executed immediately.
03Save changes: Save after selection, and it will take effect for subsequent conversations.
02
Operations that require confirmation will be displayed Wait for your confirmation to execute, click approve Execute, click reject Skip.
03
When the requirements are unclear, the agent will take the initiative to ask questions and directly choose the options given, or use Custom answer Supplementary explanation; if it is not processed temporarily, you can click Do not continue yet。
Models and costs
The model used by the agent can be selected by itself, and the conversation consumption and reserved points are displayed in real time in the panel.
01
Click the model name below the input box to open Agent model settings, the optional range is consistent with the LLM node, covering the Gemini, GPT and Claude series.
Select the model used by the agent
01Model list: The optional range is consistent with the LLM node, covering the Gemini, GPT and Claude series.
02Save model: Once selected, it will take effect for subsequent conversations.
02
Execution and points settings Displays the consumption and reserved points for this session. This number only counts the agent dialogue itself, and the points generated by the nodes are calculated separately (marked in the panel as Does not include Flows generation fee）。
Skills and Quick Recipes
Skills are special abilities that can be loaded at any time, and recipes are one-click access to common processes.
01
Click Skills Open the skills panel and browse by category or search directly. Skills are divided into built-in (Built-in) and from template (from template) two categories.
Browse and load skills
01Categories: Switch between Recommended, Favorites and My Skills.
02Search: Find skills by name or purpose.
03Skill entry: Displays the skill name, call ID and usage description. Click to load.
02
Click Use Load the skill. After loading, it will be displayed above the input box. Use skills. Click Details View instructions,Collection Collect commonly used skills,Remove skills Cancel loading.
03
Recipes are used to fix the process. Click Run recipe Afterwards, follow the prompts to select a node on the canvas or fill in the parameters to run.
Session management
Multiple conversations can be kept under the same Flows and switched, renamed or archived at any time.
01
Click Agent session View all conversations under the current Flows, with the number of messages and current status displayed for each message.
Switch the conversation under the current Flows
01New conversation: Start a new conversation, the original conversation remains in the list.
02Conversation entries: Shows the number of messages and the most recent activity.
02
Status meaning:free、Running、waiting for you to process(Answer or approval required),Completed、failed、Archived。
03
Click New conversation Start a new conversation, the original conversation will remain in the list; you can rename or archive it in the conversation menu.
2.Control from external agents (MCP)
Purpose: Without opening the website, directly operate your Flows in agents that support MCP such as Claude, ChatGPT, Codex, etc.
Connect to external agents
FlowWall provides a dedicated address, and the MCP-enabled agent can operate your Flows after being connected and authorized.
01
Open the left navigation MCP Page, copy the unique address on the page.
Copy the dedicated address and access it as a client
01Client tab: Claude, ChatGPT, OpenClaw and Hermes each have corresponding access steps.
02Exclusive address: Click to copy and fill in the connector settings of the agent.
03Other tools: Claude Code, Codex, etc. use the same address. For access methods, see the advanced instructions at the bottom of the page.
02
Fill in the address according to the steps in the client tab: Claude and ChatGPT add a custom connector in the settings; OpenClaw and Hermes write the local configuration file and log in at the terminal. Codex and other tools that support MCP use the same address and fill it in according to its own connector configuration.
03
After connecting, log in to the FlowWall account in the agent to complete authorization. Clients that require a key can be created with one click on the same page.
Common commands
Give instructions in natural language without having to remember node names or parameter formats.
01
Directly state what to do in the dialog box of the agent. Common instructions are as follows.
DATA TABLECommon commands
Command example
Execution result
Help me create a new Flows to make a 15-second short film
Create new Flows
Add an image node, use Seedream 5.0, ratio 16:9
Create nodes according to specified model and parameters
Connect the text node to the image, and then connect the image to the first frame of the video
Connect nodes in order
Run all nodes on the canvas
Run in batches and wait for results
Put these material nodes into a group and mark them in orange
Group and organize canvases
How many points do I have left? How much did that step cost?
Check balance and current consumption
Give instructions directly to the agent
01Your instructions: Describe what to do in normal conversational terms.
02Execution instructions and costs: The reply will explain what was done, how many points are expected to be consumed, and how long it will take.
03Result: After the generation is completed, a clickable finished film will be given directly in the dialogue.
Results and costs
The execution process and results will appear in both the agent dialogue and your Flows.
01
After each instruction is executed, the node will appear on Flows in real time. When performing a long task, it is recommended to open the Flows page at the same time to watch the process.
Nodes appear on Flows in real time
01Prompt word node: The screen description in the command is written as a text node.
02Picture node: Create according to the specified model and frame, and connect the prompt words.
03Video node: After the picture, the entire process is connected and can be run at any time.
02
After the generation is completed, the agent will return a clickable result link; the same product is also saved in the canvas node and Assets.
03
The estimated consumption, actual consumption and remaining balance of each step will be explained in the dialogue, and points will be automatically refunded if the generation fails.