Waku: A Native GPU-Accelerated Interface for Coding Agents

Waku is a native desktop application designed to serve as a unified interface for existing coding agent CLIs. By consolidating sessions, transcripts, and tool activity into a single GPU-accelerated window, Waku allows developers to manage multiple agents without switching between disparate terminal windows or provider-specific interfaces.

High-Performance Native Architecture

Waku is built using Rust and GPUI, the GPU-accelerated framework that powers the Zed editor. This architectural choice eliminates the overhead associated with Electron-based applications, resulting in instant launch times and smooth scrolling through extensive session transcripts.

Unified Agent Orchestration

Waku acts as a provider-neutral wrapper that normalizes various agent interfaces into a single model. It connects to agents via their strongest native interfaces, including:

  • stream-json
  • JSON-RPC
  • live events

This allows users to drive the agent CLIs they already use through a single, cohesive timeline of activity.

Integrated State Management and Rollbacks

Waku implements a "rewind" feature that synchronizes the conversation history with the actual state of the codebase. Every prompt triggers a checkpoint of the working tree under a hidden git reference. This ensures that rolling back a conversation does not just delete the chat log, but physically reverts the code to the state it existed in at that specific point in the conversation.

Local-First Design and Privacy

Waku is architected to be entirely local. All projects, sessions, transcripts, and provider IDs are stored on the user's local disk. The application contains no telemetry and does not require a cloud account, ensuring that the interaction between the developer and their agents remains private and direct.

Keyboard-Centric Workflow

The application is optimized for keyboard-driven development to minimize mouse reliance. Key shortcuts include:

  • ℓN: Start a new session
  • Enter: Queue a follow-up prompt while an agent is currently working
  • ℓEnter: Steer the agent mid-turn
  • Escape: Stop the current agent activity

Community Insights and Technical Context

Users in the Hacker News community have highlighted the growing trend of using GPUI for Rust-based GUI applications due to its performance and aesthetic appeal. Some developers noted that while native apps are compelling, the primary value lies in the ability to maintain a single timeline of every tool call, diff, and approval across multiple agents.

One user mentioned the emergence of similar tools in the ecosystem, noting that GPUI is becoming a popular choice for developers seeking high-performance alternatives to traditional web-tech stacks for desktop software.

Sources

Related

  • Project
    egoist/waku

    A fast, native desktop app that provides a unified local interface for managing multiple AI coding agents and their sessions.

  • Project
    tontinton/maki

    A high-performance, token-efficient AI coding agent with a Rust-based TUI, designed to minimize context window pollution and costs.

  • Project
    iAmCorey/kooky

    A minimal, GPU-accelerated terminal for macOS designed for AI coding, featuring one-click agent launches, workspace organization, and real-time agent activity tracking.

  • Project
    minghinmatthewlam/pi-gui

    A desktop application that provides a graphical user interface for the pi coding agent, featuring session timelines, integrated terminals, and git worktree support.

  • Project
    ShenSeanChen/waku-agent

    A local-first personal AI assistant and readable blueprint that demonstrates the core pillars of agentic loops, memory management, and LLM-Ops without using complex frameworks.