Task Agent × Visual Workflow
Use an Agent for complex work. Use a workflow for repeatable processes.
Describe the outcome and add your materials. A Task Agent plans the work, uses tools, and organizes reports, datasets, webpages, and other deliverables in your workspace; some third-party media results are delivered by link. Use a visual workflow for processes you need to run repeatedly.
New accounts receive 5,000 credits. No card required.
Split the goal into research and CSV preparation, collecting websites, social profiles, locations, and verifiable funding details.
A sub-agent compiled organizations and sources across 14 regions in research_output.md for the main task to review.Read the research and organized it into a CSV with eight named fields, keeping N/A where funding data was unavailable.Generated a download link for the CSV containing 47 organization records and completed the task.Public cases
Real tasks and deliverables
These tasks were shared publicly by their creators. Open a case to inspect the published steps, tool records, and deliverable files.
Agent benchmark selection guide
Compared 17 open-source Agent benchmarks and delivered an interactive HTML guide with test-stack recommendations.
City culture flow map
Maps 18 cities and 35 cultural routes on a rotatable 3D globe with drag, zoom, and city details.
European accelerator and incubator database
Compiled 47 accelerators and incubators into a CSV and research note covering websites, social profiles, locations, and funding details where available.
Loaded Waters VIRAL marketing video
Turned four product images into a market brief, a five-shot script, and a 15-second vertical video generated with Sora.
2025 Fed rate-cut cycle research report
Based on public material available through December 2025, delivered a cited report spanning eight themes and four rate-cut cycles.
AI Agent memory technology review
Compared five core memory benchmarks and seven open-source approaches, with scenario-based selection and phased adoption guidance.
Platform at scale
across text · image · video · audio
across 17 functional categories
for files · search · terminals · sub-tasks
Task Agent · visual workflow
Two ways to work
Open-ended work and repeatable processes
Use a Task Agent when the path may change and the work needs judgment. Build a visual workflow when the steps are known and need to run repeatedly. Both can connect models, files, data, and external tools.
Task Agent
Give a goal and materials; the Agent plans, executes, and corrects itself. Built for complex work without a fixed path.
- Review messages, actions, observations, and tool results by cycle
- Use files, knowledge bases, workflows, MCP, and sub-agents
- Add context during a task and receive completion notices
Visual Workflow
Define inputs, processing, and outputs with nodes for processes with known, repeatable steps.
- Drag to connect models, data, media and output nodes
- Official & community templates, or start blank
- Run on a schedule, keep records, and share when needed
Agents can invoke configured workflows and MCP tools. Workflows can also be added to an MCP Server for compatible clients to call.
Collaboration across your organization
Use organization-shared files directly in Agents and workflows
Admins can manage members and nested department structures. Share files and folders with selected members, departments, or the whole organization, with view-only or collaboration access. Authorized shared files can be used in Agents and workflows, and Agent-generated files can be saved back to shared folders where you have collaboration access.
Organizations, departments, and members
Manage members, invitations, and multi-level departments so sharing follows the way your team is structured.
Share materials by scope
Share files and folders with selected members, departments, or the whole organization, with view-only or collaboration access.
Use shared files in Agents and workflows
Use shared files as Agent reference material or workflow inputs, then save Agent-generated files back to shared folders where you have collaboration access.
Choosing a tool
VectorVein, Codex, or Claude Code?
Choose based on both the final deliverable and what your team needs to share over time. VectorVein lets teams use organization-shared files in Agents and workflows, then save outputs back to shared folders; Codex and Claude Code center collaboration on repositories and software-development processes.
| Dimension | VectorVein | Codex | Claude Code |
|---|---|---|---|
| Primary work | Research, content, data processing, and other multi-step tasks with a defined deliverable. | Build, debug, refactor, and review code in a software repository, and run tests. | Build, debug, refactor, and review code in a codebase, and run tests. |
| Primary context | Webpages, Office and PDF files, datasets, knowledge bases, media, and external tools. | Code repositories, terminals, IDEs, Git, and development tools. | Codebases, terminals, IDEs, Git, and development tools. |
| Organization and team collaboration | Built-in organization, department, member, and invitation management. Share files and folders with members, departments, or the whole organization, set view-only or collaboration access, and use shared files directly in Agents and workflows. | Team collaboration mainly centers on repositories, shared development guidance, code review, pull requests, and related engineering processes. ChatGPT Business and Enterprise workspaces also provide admin controls. | Team collaboration mainly centers on repositories, project settings, GitHub workflows, pull requests, and related engineering processes. It supports team-shared settings and organization-managed configuration. |
| When to choose it | Your main deliverable is a report, dataset, webpage, or media result, or your organization needs to reuse shared files and standardize repeatable processes as workflows. | Your main deliverable is code and you prefer OpenAI models, or want to work through the desktop app, IDE, CLI, cloud tasks in the web app, or SDK. | Your main deliverable is code and you prefer Claude models, or want to work through the terminal, IDE, desktop app, browser, GitHub Actions, or Agent SDK. |
| Typical outputs | Reports, datasets, webpages, media results, workspace files, and reusable workflows. | Code changes, diffs, test or command results, code-review feedback, commits, and pull requests. | Code changes, diffs, test or command results, code-review feedback, commits, and pull requests. |
Use them togetherThey can divide the work: use Codex or Claude Code to build and maintain code, and VectorVein to run and reuse business processes. Claude Code can call published workflows through VectorVein's current SSE MCP endpoint. Codex remote MCP currently requires Streamable HTTP, so use a REST API or an adapter for now.
Features and access options can change. See the official documentation: CodexCodex MCPClaude CodeClaude Code SettingsClaude Code MCP
Core capabilities
Run code, work with files, and deliver usable outputs
Review task steps and tool calls, then find reports, webpages, datasets, and other deliverable files in the workspace. Some third-party media results are delivered by link.
Process files and run analysis
Agents can run Bash and Python and read or write files in an isolated cloud environment that shuts down when idle.
Move complex work forward in parallel
A main Agent can split work into sub-tasks so research, data processing, and report writing can proceed in parallel.
Retrieve material with source file paths
Connect Office documents, PDFs, and cloud files, then retrieve relevant content with its original file path for review and citation.
Run by time or task status
Schedule runs or start follow-up tasks and workflows when a task changes state.
Reuse steps that have been verified
Agents can invoke configured workflows and external tools. Repeatable steps can be organized as visual workflows.
Bring your own API key
Use your own API keys without spending platform credits on model calls
Calls made with your own OpenAI, Azure, Anthropic, Vertex, or Bedrock keys do not use platform credits; your model provider bills those calls under its own terms. You can choose whether to enable platform-key fallback, and fallback calls use platform credits.
- Choose your own key or a platform key for each model
- Configure your own keys and platform-key fallback separately
- See an itemized breakdown of credit usage for model calls and workflow runs
Ways to access VectorVein
Start work from the devices and systems you use
Start tasks from the web workspace, browser extension, supported mobile apps, or REST APIs. Workflows can also be published as web apps or called by compatible MCP clients.
Browser extension
Send selected text, links, or page content to enabled workflows from Chrome, Firefox, or Edge.
iOS App
Available in selected App Store regions for viewing and running tasks and workflows on iPhone.
Open in the App StoreDeveloper API
Use an API key to manage and run workflows and Task Agents through REST APIs.
Manage API keysConnect and publish with MCP
Connect MCP services to Task Agents, or expose selected workflows through a configured MCP Server for compatible clients.
Publish as a VApp web application
Publish a workflow as a web app and control access with keys, usage limits, credit caps, and expiration dates.
Create a VAppHow does pricing work?
Creating and configuring Task Agents and editing workflows do not use credits. Runs consume credits based on the platform models and paid tools used. Calls made with your own third-party model keys do not use platform credits; your model provider bills those calls under its own terms. View model credit pricing
Who is VectorVein for?
VectorVein fits research, content, operations, and data work that involves finding sources, processing files, using multiple tools, and keeping reviewable steps and deliverables. Fixed processes can be built as visual workflows. For a one-off short answer, a standard chat product may be simpler.
How is VectorVein different from Codex and Claude Code?
Consider both the deliverable and what your team needs to reuse over time: choose VectorVein for reports, datasets, webpages, or media, or when your organization needs to reuse shared files and standardize repeatable processes as workflows; choose Codex or Claude Code for code changes, tests, and review around a codebase. They can also be used together. See the full comparison
Are my tasks and workspace files public by default?
No. New tasks are not shared and do not appear in the community by default. When you explicitly enable sharing in task details, anyone with the link can view the task description, execution records, tool calls, and workspace files. Enabling Public also lists the task in the community. Sharing a parent task also makes its subtasks accessible by link, so review and remove sensitive content before sharing.
Does using my own model API key consume platform credits?
Successful third-party model calls made with your own key do not use platform credits; your model provider bills you directly. You can choose to use only your own key or fall back to a platform key when it is unavailable. Only fallback calls use platform credits.
Can I try for free?
Yes. New accounts receive 5,000 credits and can try Task Agents and workflows without adding a payment method first.
Is there an open-source edition?
Yes. Download the open-source desktop edition from GitHub Releases and run it locally. It focuses on visual workflows; Task Agents are currently available on the hosted VectorVein platform.
Start with a task you need done.
Describe the outcome and add your materials. New accounts receive 5,000 credits with no payment method required.
Create a task for free