AgentOp is a browser-based platform for building, testing, and exporting private AI agents that run open-source LLMs entirely on your device — no servers, no install, no data leaving your tab.
What is AgentOp?
AgentOp is a browser-based platform that lets you build private AI agents using local open-source language models such as Llama 3.2, Qwen, Gemma, Phi, and DeepSeek — all running completely on your device via WebGPU (with CPU fallback via WebAssembly). You write agent logic as Python tools executed in-browser by Pyodide, test them instantly against a local model, and export the entire agent as a single standalone HTML file that includes the model loader, Python runtime, UI, and encryption. This file runs offline on any modern browser without any installation or account. AgentOp is free to start — no credit card required.
Key Features
- On-device inference — Models are downloaded once, cached in your browser, and inference runs on your CPU/GPU via WebGPU or WebAssembly. Your prompts and data never leave the page when using local models.
- Python tool support — Define tool schemas and handlers in Python using Pyodide 0.29, with grammar-constrained (GBNF) function calling that works identically across local and cloud models.
- Single-file export — The agent is packaged as one standalone HTML file that includes everything needed to run offline without internet, install, or account. The file never expires and can be emailed, embedded, or dropped on a USB stick.
- Cloud model integration — Optionally connect your own OpenAI or Anthropic API key for GPT-4 or Claude. Keys are encrypted client-side with AES-256 and never sent to AgentOp servers. You can mix local and cloud calls: cheap drafts locally, hard calls in the cloud.
- AI agent generator — Describe your agent in plain English and AI writes the complete agent (HTML, Python tools, prompts, UI) in seconds. Your API key stays in your browser and makes only one call to the LLM — AgentOp never sees the key.
- Marketplace of agents and templates — Browse public agents and templates to fork and customize, or publish your own agent to the marketplace.
- Privacy by architecture — When using local models, your data never leaves your device. For cloud models, your key is encrypted client-side and never stored by AgentOp. No account or sign-up required for local use.
Who should use AgentOp?
- Developers — Prototype AI agents in Python without standing up any infrastructure. Test against a local model instantly without an API key, then export a shippable file.
- Privacy-first teams — Keep sensitive documents and conversations on-device. Nothing is sent to any server when using local models.
- Educators — Hand out a single HTML file that runs offline on any machine, no install or account needed. Students can interact with a real LLM in a controlled environment.
- Tinkerers — Fork, remix, and break things with zero setup cost. The entire agent is in one portable file.
Use cases
- Build a custom research assistant — Use Python tools to search local files, fetch web content, or call APIs, then package the agent as a single file for your team.
- Create an offline code reviewer — Write Python tools that analyze code, and export the agent for use without internet access.
- Deploy a private client Q&A bot — Keep customer data on-device by running a local model, with the option to escalate to a cloud model for complex queries using your own API key.
- Generate a template from a description — Describe your idea in plain English, and AI writes the entire agent — you land on a working agent ready to customize or export.
How does AgentOp work?
- Pick a template from the curated collection or describe your idea to generate one via AI.
- Edit and test Python tools locally — the model runs in your browser, so no API key is needed for testing.
- Export a standalone HTML file that includes everything: model loader, Python runtime, UI, and encryption. The file runs offline on any browser without setup.
Pricing
AgentOp is free to start. No credit card required. Local model inference is free. Cloud model usage requires your own API key.
FAQ
Is it really running locally?
Yes. The model downloads once, caches in your browser, and inference runs on your CPU/GPU. Open your network tab — it is empty during a chat. There is no server involved.
Do I need a GPU?
No. A GPU is not required. The engine falls back to your CPU via WebAssembly if WebGPU is unavailable. In Chrome and Edge, the model can run GPU-accelerated via WebGPU for faster inference. Smaller models run fine on a laptop either way.
What about my data and prompts?
They never leave the page when you use a local model. If you bring your own API key for a cloud model, the key is encrypted client-side with AES-256 and never stored by AgentOp.
Can I still use GPT-4 or Claude?
Yes — bring your own key. It is encrypted client-side with AES-256 and never sent to our servers. You can mix local and cloud calls.
What does sharing a file give the recipient?
The entire application. They open one HTML file and it runs on their machine, offline — no install, no account required. The file never expires.
