Waza
Switch between leading AI models, operate files, run code, and visualize data — a hands-on agentic AI workspace built to move work forward.
How Waza changes the way you work
Waza was built from a hands-on analysis of what kind of AI agent is actually needed to transform how work gets done.
Beyond standard file and command operations, its defining feature is two-way communication between chat and artifacts. AI-generated documents and tables are fully interactive — actions taken inside an artifact feed back directly into the chat.
The biggest difference from other AI agents is that Waza is deliberately non-autonomous. Autonomous agents keep running on their own, which can drive costs through the ceiling and carries the risk of processing continuing without visibility into what's happening. Waza is designed for human judgment and intervention, so you can control costs while using it with confidence.
There are other cost-control mechanisms too. Many agents rely heavily on screenshot recognition for browser operations, which increases API calls and drives up costs. Waza avoids this by using DOM operations as the foundation, and supports switching between multiple models so you can optimize costs based on your use case.
Prompt management is also simple. There are no complex concepts like "skills" found in other tools — just save frequently used prompts and recall them with a single click. Anyone can get started right away without IT knowledge. It also includes mechanisms that make hallucinations less likely in calculations and conditional judgment scenarios.
| Autonomous AI agents | Waza | |
|---|---|---|
| Operating costs | Can spiral out of control | Human-controlled, predictable |
| Browser operations | Screenshot-dependent, high cost | DOM-based, stable and low-cost |
| Transparency | Hard to see what's happening | Step-by-step verification and intervention |
| Prompt reuse | Complex skill registration | Save and recall with one click |
| Calculation accuracy | Prone to hallucinations | Code execution ensures precision |
| File operations | Supported | Equivalent capabilities |
What it looks like in practice
Main features
Conversation & AI
- Give instructions to AI in natural language
- See responses in real time (streaming)
- Stop generation at any time
- Switch between AI models
File Operations
- Read and write file contents
- Browse folder contents
- Search files by keyword (Japanese supported)
- Move or copy multiple files at once
Command Execution
- Run shell commands (with a confirmation step before execution)
Attachments
- Paste an image for AI to analyze
- Share CSV or text files for AI to process
- Drag and drop HTML files to render them
Generated Content (Side Panel)
- HTML: Render and interact with web pages
- Code: Syntax-highlighted program output
- Table: Edit cells, add/delete rows, export CSV
- Chart: Interactive data visualization
Task Automation
- Save successful workflows as reusable "skills"
- Re-run saved skills instantly next time
- Approve, discard, or delete skills at any time
Security
- File writes, moves, and command execution always show a confirmation dialog
- Nothing runs without your explicit approval
Python Execution
- Run numeric calculations, aggregations, and statistics with actual Python code
- Returns executed results, not guesses — prevents hallucinations
- Results displayed in the artifact panel
Latest Updates
Capabilities grow continuously. New features are added based on real workflow needs. If you want to confirm what's currently possible or explore a specific use case, feel free to reach out.
Talk to us about Waza
Curious how it fits your workflow? We're happy to explore it together.