Kanna: A Tactile Intelligent Workflow Tool Based on Native Agent CLI

Kanna is an innovative tactile interactive intelligent workflow tool. By integrating native Agent CLIs like OpenAI Codex and Claude Code, it introduces tactile interaction modes such as gesture recognition, tactile feedback, and spatial interaction. As a meta-layer tool for orchestrating Agent capabilities, it redefines the interaction paradigm of human-computer collaboration and explores a more natural and intuitive AI Agent user experience.

As AI Agents evolve from concept to practical use, native Agent tools like OpenAI Codex CLI and Anthropic Claude Code provide powerful intelligent programming assistants, but their interaction is limited to traditional command lines. Kanna originated from the idea of combining Agent CLI with natural and intuitive interaction. Its core pursuit is "tactile" interaction—controlling AI workflows through physical gestures and tactile feedback, breaking the single dimension of keyboard input, and creating a brand-new human-computer collaboration experience.

Kanna's core features include: 1. Multi-Agent CLI Integration: As a unified orchestration layer, it seamlessly integrates mainstream Agent CLIs, allowing users to switch flexibly; 2. Tactile Interaction Layer: Includes gesture recognition (mapped to commands), tactile feedback (operation confirmation), and spatial interaction (3D code navigation); 3. Workflow Orchestration Engine: A visual designer decomposes tasks into reusable step sequences; 4. Context-Aware System: Maintains information such as codebase status and conversation history to assist Agent decision-making.

Applicable to multiple development scenarios: Immersive code review (navigate code with gestures, mark issues to trigger fixes); Multi-file refactoring (grab files in 3D space to show relationships, trigger cross-file refactoring); Real-time code collaboration (team shares the interactive interface, multiple developers collaborate via tactile devices); Demonstration and teaching (intuitively show Agent processes with gestures, lowering the learning threshold).

Faces multiple challenges: Low latency requirements (Agent calls, gesture recognition, and feedback need millisecond-level coordination); Cross-platform compatibility (a unified abstraction layer adapts to different Agent APIs and authentication); Gesture design (balance expressiveness and simplicity to avoid accidental touches); Hardware ecosystem dependence (supports a wide range of hardware from ordinary cameras to professional VR/AR devices).

Kanna is positioned as a "meta-layer" tool: For Agent CLIs, it is a caller and orchestrator; For IDEs, it can run in parallel or be embedded to provide additional interaction dimensions; For users, it is a unified entry point. It does not compete with existing tools but enhances their capabilities through innovative interaction, having unique niche value.

If realized, it may bring: Evolution of interaction paradigm (from "telling" to "showing" the Agent what to do); Improved accessibility (gesture/tactile interaction benefits users with disabilities, democratizing access to AI Agents); Preparation for spatial computing (adapts to 3D interaction, aligning with device trends like Apple Vision Pro); Prelude to multimodal AI (provides practice for future multimodal Agent inputs).

Kanna is in the early open-source stage, with the GitHub repository providing basic code structure and documentation. Developers can participate in: Exploring Agent CLI integration interfaces; Contributing gesture recognition or hardware adaptation code; Designing workflow templates; Improving documentation and tutorials.