Flushnet AI Runner
The AI automation Agent that runs your computer and browser.
Run browser, terminal, and AI tasks across your machines and remote nodes from one place.
The fastest way to connect a machine, launch a task, watch execution, and get the result from one operator layer. Use your own machines or secure sandboxes. Run tasks in parallel. Control everything from the dashboard, API, or AI assistant.
Start free Install agent View docsStart with AI Runner Chat, Chrome extension, local runtime, or sandbox
For ChatGPT control, first link the AI Runner chat to your Flushnet account with a dashboard access code. Then choose how tasks will run: Chrome extension, local desktop runtime, or sandbox/browser VM mode.
- Log in or create your Flushnet account.
- Open Dashboard and then Generate Access Code. Free-user access codes expire after 30 minutes; Pro and Power codes do not expire while the plan is active.
- Open AI Runner Chat in ChatGPT and paste the access code when asked.
- Choose a runtime path: install the Chrome extension for browser-based control, install the local runtime for native desktop/file/terminal control, or use sandbox/browser VM mode.
- Ask the chat to list nodes, confirm the right node or sandbox is online, then run your first task.
Free users can use sandbox/browser VM mode. Native/local laptop control, file tools, and terminal access require Pro, Power, or admin access.
Open AI Runner Chat · Generate Access Code · Install runtime · See onboarding examples
AI Runner turns requests into real actions across desktop, browser, mobile, and connected devices.
Use one AI operator layer to organize files, automate workflows, fill forms, test apps, run programs, manage email and calendars, compare products, create reports, place calls, chat on Telegram or WhatsApp, and control IoT devices.
Built for practical execution: browser automation, terminal tasks, multimodal AI, mobile camera and voice input, call center and conferencing flows, sandbox and virtual machines, RAG memory, third-party app integrations, and support for ChatGPT, Claude, Gemini, Ollama, Hugging Face, and llama.cpp.
Some of the tasks you can ask AI Runner to do
- Organize my computer files
- Automate and run this workflow
- Fill in this online form for me
- Build and test a mobile app
- Run a program on my computer
- Check the last emails from my boss and write a reply with the latest work updates
- Test and debug my app
- Write and post my new event to my Facebook page
- Create a photo gallery of my personal photos
- Book a restaurant reservation
- Search and compare prices for this product
- Search in “My Documents” for statements and write a full report in PDF format
- Schedule an appointment on my calendar
- Dial this telephone number
- Read my Telegram and WhatsApp messages
- Use the mobile phone app or desktop app with live audio voice to open and write an MS Word document
- Use the mobile camera app to send photos and videos to this chat
- Use the mobile app to show my current AI Runner tasks
- Chat with AI Runner from Telegram and WhatsApp in text and audio chat
- Enable all my home IoT smart devices such as sockets, cameras, and robots
More features include
- Windows and Mac installers
- Support for OpenAI ChatGPT models 5.4, 5.3, and 5.2, Anthropic Claude, and Google Gemini
- Full integration with OpenClaw, open-source, and third-party AI agents
- Integration with Ollama, Hugging Face, and llama.cpp
- Full multimodal support and custom multimodal AI models
- Task and workflow optimization
- AI multi-channel support including IoT, WhatsApp, Telegram audio, phone, call center, video call, multi-user conference, and more
- Third-party and open-source app integration
- Sandbox and virtual machine mode
- RAG and persistent long-term AI memory support
- Login verification via access code and 2FA authentication
How it works
- Connect a machine or open a sandbox
- Launch a task from dashboard, API, or AI assistant
- Watch execution and collect the result
Browser tasks
Open a site, click, type, capture a screenshot, or run repeatable browser automation on a connected machine.
Terminal and file tasks
Run commands, inspect output, move files, and validate changes on a real machine or in a safer sandbox first.
AI and batch jobs
Launch compute workflows, repo jobs, or repeated tasks across one node or many machines from the same control layer.
Why it is different
- Your own machines plus remote nodes
- Live control and visibility
- Parallel execution from one place
Proof of workflow
Typical flow: install one agent → confirm the node is online → run a safe command → scale the same task across more machines.
API transport
Direct API access via /api. Supports v2 and v3.
Start here
Beginner, professional, and enterprise step-by-step examples are in the Help page, including custom GPT, sandbox-first, native machine, and rollout workflows.
Pricing
Start free, upgrade for native machine control, larger queues, and broader orchestration when you are ready.
Why AI Runner is different
Most AI tools stop at answering, coding, or controlling one narrow environment. Flushnet AI Runner is built as an execution layer: it connects AI reasoning with real machines, browser sessions, workflows, approvals, and operational visibility.
| Category | Flushnet AI Runner | AI Chat Assistants | Browser Automation Tools | Developer Agent Platforms |
|---|---|---|---|---|
| Core purpose | Execute real tasks across connected environments | Answer questions and assist with content | Automate websites and browser flows | Help build, test, and modify software |
| Execution scope | Desktop, browser, files, scripts, workflows, and remote nodes | Mostly chat-based guidance | Mostly browser-only | Mostly coding environments |
| Operator control | Dashboard, API, AI chat, and connected runtimes | Conversation only | Script or test runner | Repo, terminal, or cloud workspace |
| Task handling | Adaptive execution with validation and retry logic | Suggests steps; user executes | Requires predefined automation logic | Strong for code tasks; limited outside dev workflows |
| Machine flexibility | Run locally, remotely, or in isolated environments | Usually no machine execution | Usually tied to browser sessions | Usually tied to sandboxes or developer workspaces |
| Workflow depth | Repeatable, scheduled, parallel, and approval-based operations | Limited | Requires custom engineering | Usually developer-focused |
| Best fit | Operators who want AI to complete real work | Users who want advice or drafting help | Teams automating web interactions | Developers building or maintaining software |
Flushnet AI Runner focuses on the outcome: AI that can safely move from instruction to action, then prove what changed.
Detailed platform comparison
| Capability area | AI Runner | Claude Desktop | OpenClaw | OpenHands |
|---|---|---|---|---|
| Primary role | AI operator for real task execution | AI desktop assistant | Automation framework | Software engineering agent |
| Work environment | Desktop, browser, files, workflows, and connected runtimes | Mostly assistant-led desktop workflows | Automation and integration environments | Code, repos, terminals, and sandboxes |
| Real-world task execution | Strong | Partial | Partial | Partial |
| Browser-based work | Strong | Partial | Strong | Strong |
| Technical and developer workflows | Strong | Partial | Partial | Strong |
| Multi-step automation | Strong | Partial | Strong | Strong |
| Remote or isolated execution | Strong | Limited | Strong | Partial |
| Operational visibility | Strong | Limited | Partial | Partial |
| Human-guided control | Strong | Partial | Partial | Partial |
| Extensibility | Broad platform access | Ecosystem-based | Open automation stack | Developer-focused |
| Best fit | Operators who want AI to complete and verify work | Users who want assistant support | Teams building automation stacks | Developers automating software work |