Sign in required
Sign inVisFlow
VisFlow is a browser agent that can learn from a single demonstration. Tell it what you need or demonstrate the workflow once. It observes the page, decides what to do next, and turns repetitive browser work into reusable execution knowledge.
You may have run into this too:
Repetitive work such as expense claims, timesheets, and asset registration looks simple, but consumes time every day—and many internal systems still lack stable APIs.
A minor redesign or a moved dialog can break a rigid, preconfigured workflow.
Solutions that rely only on page structure struggle with Canvas, custom-rendered controls, and complex interactions.
A general agent may understand the goal but not the site-specific workflow or company rules, leading to repeated attempts at critical steps and an unfinished task.
What is missing is usually not the goal, but an understanding of the visible page state and the operational knowledge needed to act.
Visual-first execution:Understand the page users actually see, then use page structure to supplement long pages and structured information.
Learning from demonstrations:Demonstrate the task naturally once. VisFlow captures reusable operational knowledge and applies it flexibly to similar tasks.
Find information on the web:Find web information such as flights and traffic restrictions.
Summarize page content:Summarize the main content of the current page.
Find and interact with content:Find, save, or play specified content.
Complete web forms:Complete registration, appointment, and other web forms.
Handle repetitive workflows:Handle repetitive office and back-office workflows.
VisFlow does more than explain a workflow: it can click and fill forms for you. You can pause or take over at any time, and important actions such as submitting, paying, or deleting require confirmation.
Rule-based workflow tools｜Effective when workflows are predictable: Workflows must be configured and maintained; Operational knowledge is locked in scripts; UI changes often require rework.
Browser-interface solutions｜Flexible on conventional web pages: Depend on an upstream agent for reasoning and planning; Work best with well-structured, automation-friendly pages; Complex interfaces and specialized workflows may require repeated exploration.
VisFlow｜Visual execution + demonstration learning: An integrated loop for perception, planning, and execution; Learns from natural demonstrations without specially authored recordings; Reuses experience for similar tasks instead of replaying rigid macros.
Visual-first, structure-assisted: First understand the final rendered view, then use page structure to supplement long pages and structured information. This covers visual interfaces such as Canvas and custom controls while retaining support for tables and forms, with cross-checks before high-risk actions.
Open integration:VisFlow provides MCP and Skill integrations for general agents including Codex, Claude Code, WisCode, and Cursor. The general agent understands the goal, decomposes the task, and coordinates tools, while VisFlow handles the web steps that are difficult to execute reliably.
01 · Understand and plan｜The general agent coordinates the whole task: Understand complex goals and context; Break down cross-tool, cross-page tasks; Decide when to invoke VisFlow.
02 · MCP / Skill Integration｜Bring reliable web execution into your workflow: Works with general agents including Codex, Claude Code, WisCode, and Cursor; Connects to existing workflows through standard interfaces; No need to rebuild web automation for every agent.
03 · Execute reliably｜VisFlow handles the web steps: Understand the rendered page and its visible state; Reuse operational knowledge captured from demonstrations; Return results so orchestration can continue.
27 steps｜Completed in about one minute: In an internal ERP system, VisFlow completed an end-to-end IT asset request spanning up to 27 steps, so new employees no longer needed to consult an operating manual.
600+ requests｜Validated at batch scale: In stress testing, VisFlow submitted more than 600 asset requests and supported one-click batch rejection, validating the efficiency potential of workflow-level automation.
First-pass success｜From unsuccessful exploration to first-pass success: On an unfamiliar internal system, six minutes of unguided exploration failed. After a two-minute human demonstration, the next run succeeded on the first attempt in about one minute.
Enterprise workflow automation: Automate standard workflows across office systems, expenses, timesheets, asset requests, IT services, and internal forms.
Workflow reproduction and knowledge reuse: Capture standard procedures as task examples for reproduction, onboarding, and related workflows.
Batch data entry and processing: Handle data entry, status updates, batch submissions, and back-office maintenance.
Intelligent web search and form completion: Reduce repetitive typing and clicking for registration, data entry, appointment requests, surveys, and other web forms.
Enterprise security and compliance:VisFlow establishes safeguards across data, actions, and permissions, bringing AI into business workflows safely, compliantly, and under human control.
Sensitive-data masking: Mask sensitive fields in workflow records and page data to reduce risk when capturing and reusing examples.
Confirmation for high-risk actions: Validate intent and require human confirmation before critical actions such as submitting, approving, deleting, paying, or publishing.
Operation within existing permissions: Operate within the user’s existing browser session and permissions, without privilege escalation or unrestricted screen access.
Private model integration: Connect to an enterprise LLM gateway or designated model so data handling and inference remain controllable.
Install the extension: Install VisFlow in Chrome or Edge.Open the Chrome Web Store
Open the target page: Open the target website. Sign in first if the task requires an account.
Enter a task: Open VisFlow and describe what you need in one sentence.