Honest writing about what we're making, what we've shipped, what's broken, and what we're learning while building a gesture-based terminal multiplexer for the AI-agent era.
A full audit, accurate as of May 2026 — every shipped feature, every paused feature, every abandoned feature.
Deep dives: where gmux came from, how each implementation evolved, where it sits in the ecosystem.
The three pillars — gestures, agent management, visual oversight — that turn a fleet of AI agents from chaos into a workspace you can see. Big SVG visuals throughout. Demo CTAs at the top.
Best place to start if you've never seen gmux. Read overview Research · 02 · May 13 · ~10 minSix implementations, three pillars, one stack. From the April prototype to the Tauri desktop app to gmux-brain memory. With a layered architecture diagram and the camera-broker fix that solved gestures.
For developers — covers the Tauri/PTY architecture in detail. Read the stack Research · 03 · May 13 · ~9 minThe annoying-to-solve problem. The state-detection fix that made management work. The Wayland trap. The port collision with aria-phone. The installer freeze. The qalcode2 overlap realisation. What gets shipped next.
Personal-tone story. Best for blog / HN audiences. Read the devlog Research · 04 · May 13 · ~8 minThree layers of AI tooling: orchestration (Multica, Linear), execution (qalcode2, opencode, DeepSeek-TUI), interaction (gmux). Most tools cluster around the first two. The third is empty — that's the moat.
For pitch context. The three-layer diagram is the takeaway. Read the positioningShorter pieces — one idea each. The pitch, the architecture, the gesture vocabulary, the phone, the memory layer, and the roadmap.
More writing arrives as we ship — the next piece will be a launch note when the Tauri app passes its five-criterion install gate. Sign up to find out first.