Hanzo Operative: Computer Use for AI Agents
Introducing Hanzo Operative — enabling AI agents to use computers like humans: screen capture, mouse, keyboard, and application interaction.
AI agents can write code, search the web, and process documents. But most software was built for humans using screens, mice, and keyboards. Until agents can interact with arbitrary software through its UI, they are limited to API-accessible tools.
Today we are launching Hanzo Operative: computer use infrastructure for AI agents.
What Operative Does
Operative gives AI agents the ability to interact with any computer application:
- Screen capture: Real-time screen observation with element detection
- Mouse control: Click, drag, scroll, hover with pixel precision
- Keyboard input: Type text, use keyboard shortcuts, fill forms
- Application interaction: Navigate menus, switch windows, manage tabs
The agent sees the screen, reasons about what to do, and takes action — the same loop a human follows.
Architecture
Operative runs in a sandboxed environment:
- Isolated virtual display per agent session
- Network policies to control what the agent can access
- Action logging and audit trail
- Session recording for review and debugging
- Timeout and resource limits
Use Cases
Legacy System Integration
Connect AI workflows to software that has no API:
- Enterprise applications with only a GUI
- Desktop software (accounting, design, proprietary tools)
- Government portals and forms
- Legacy web applications
Testing and QA
AI-powered end-to-end testing:
- Visual regression testing
- User flow validation
- Cross-browser testing
- Accessibility auditing
Data Entry and Processing
Automate repetitive UI tasks:
- Form filling across multiple systems
- Data migration between applications
- Report generation from multiple sources
- Document processing workflows
Safety
Computer use requires careful safety boundaries:
- Configurable action allowlists (which applications, which actions)
- Human-in-the-loop confirmation for sensitive actions
- Network isolation to prevent unintended external access
- Complete audit log of every action taken
Pricing
- Free: 100 minutes/month
- Pro: $49/month — 1,000 minutes, all features
- Enterprise: Custom minutes, dedicated infrastructure
Hanzo AI is Techstars '17.
Read more
Hanzo Dev: AI Coding Agent for Your Terminal
Hanzo Dev launches — an AI coding agent that lives in your terminal, understands your codebase, and writes production-quality code.
Zen VL: Vision-Language Models with Function Calling
Zen VL is a family of vision-language models at 4B, 8B, and 30B -- each with instruct and agent variants -- supporting OCR in 32 languages, GUI navigation, spatial grounding, and native function calling with visual context.
Hanzo Brain: Local-First Knowledge Graph for Agents
A single SQLite file your agent reads from every runtime. Zero LLM round-trips for graph ops, FTS5 plus vector plus typed-edge extraction at write time.