Native Swift Mac app
David started with a native SwiftUI/AppKit foundation so the always-on host can stay lightweight, responsive, and close to macOS.
Features
Native Mac control for AI workers.
David gives Claude Code and Codex workers a home base: lightweight Mac control, project memory, loop contracts, CodeGraph context, approvals, and self-improvement that stays governed by you.
Core strengths
The pieces that make David more than another prompt window.
Durable workspace memory
Workers can leave checkpoints, project notes, artifacts, and catch-up reports that help the next session restart from context instead of a blank prompt.
Loop contracts
Launch, Maintenance, Security, Improve, Animation, and Dream Catch-Up loops can carry termination rules, verification gates, fallback behavior, artifacts, and approvals.
CodeGraph context
When a repo has graph data, David can prepare search, file, and impact queries so workers inspect the right code before wandering through the tree.
Governed self-improvement
Agents can propose improvements, but David keeps the evidence trail: policy, version, approval, applied log, and rollback path.
Runtime choice
David is being built to supervise Claude Code and Codex jobs with workspace defaults, per-worker runtime controls, model choices, and scratch chats.
Why native matters
A Mac app can be a quiet control room.
David can keep the coordination layer separate from worker contexts, park idle work, resume from durable state, and stay polite on laptops. The goal is not a bigger chat box. It is a calm local place where long-running AI work can be supervised.
Small host, heavier work delegated to workers
Approvals and Needs You gates stay visible
Direct-download release path for a closed-source app
macOS-native direction for notifications, Keychain, Shortcuts, and system-aware behavior
Current surfaces
Built around the real workflow.
Early access
Build with workers you can actually supervise.
David is in release-candidate testing as a direct-download Mac app for teams and solo builders who want loop engineering, memory, and approvals around their AI coding workers.