Where gmux sits in the ecosystem.

The AI developer tooling space is moving fast enough that categories are still forming. Tools that look similar on the surface — "AI terminal things" — are actually solving different problems at different layers of the stack. Understanding where gmux fits requires being precise about what layer each tool operates on.

The three layers

When you're running AI coding agents, three distinct problems need solving:

Most tools in the space cover Layer 2 well. Some cover Layer 1. Almost nothing covers Layer 3, because Layer 3 only becomes a problem when you have enough agents running that interaction becomes the bottleneck. That threshold is higher than most people have reached yet, but it's coming.

Layer 1 · orchestration tools

Multica (27k ⭐)

Multica's tagline is "your next 10 hires won't be human." It's a project management layer where AI agents are first-class assignees — they appear in task boards alongside humans, pick up tickets, report blockers, and mark tasks complete. Think Linear or Jira, but agents are actual workers.

The interface is a web dashboard. You create a task, assign it to an AI agent, and the agent's daemon picks it up and executes it. Multica tracks what's in progress, what's blocked, what's done.

Multica auto-detects supported runtimes on your PATH at first run. Its current list includes: Claude Code, Codex, Cursor, GitHub Copilot CLI, Gemini CLI, Hermes, Kimi, Kiro CLI, OpenCode, OpenClaw, Pi. OpenCode is on that list — which means qalcode2 (a fork of OpenCode with the same CLI surface) is detected automatically.

Relationship to gmux: Multica is the task source; gmux is the terminal where tasks execute. When Multica assigns a task to a qalcode2 agent, that agent opens in a tmux window. gmux shows you the state of that window and every other window alongside it. The stack:

Multica  →  assigns task to qalcode2 agent
               ↓  agent runs in tmux window
           gmux  →  shows state, handles gesture/voice interaction

These tools don't compete. Multica answers "what's queued." gmux answers "what needs my attention right now."

Hermes Agent (145k ⭐)

Hermes is a self-improving agent CLI from Nous Research. It runs persistent, cross-session learning — creating skills from experience, improving those skills, building a user model. It has its own messaging gateway (Signal, WhatsApp, Slack, Telegram, Email all from one instance), parallel subagents, and a cron scheduler for unattended automation.

Hermes is also one of Multica's supported runtimes, meaning Multica can assign tasks to Hermes agents the same way it assigns them to qalcode2.

Relationship to gmux: Hermes is another agent that would run inside a gmux-managed tmux pane. gmux would show its state (working/waiting/permission), handle gesture selection, route voice commands to it. The Hermes agent handles the intelligence and persistence; gmux handles the visual experience of managing it alongside other agents.

Hermes's skills system (based on the agentskills.io standard, using .claude/skills/ conventions) is compatible with the SKILL.md convention used in graphify. Cross-pollination is possible — skills written for gmux's context could be invoked by Hermes agents running inside gmux panes.

Layer 2 · execution tools

qalcode2

qalcode2 is the primary execution runtime in the gmux ecosystem — the AI coding agent that runs inside each tmux pane. It's a fork of opencode with non-AVX hardware support and additional features including the HTTP API that gmux reads for state.

Every gmux feature that surfaces AI state (the status bar, the sidebar state dots, the todo progress) is reading data that qalcode2 generates. gmux doesn't generate state — it aggregates it from multiple qalcode2 instances and presents it in one view.

This is the key distinction: qalcode2 is an excellent single-agent tool. gmux is what you add when you're running ten of them.

opencode / Claude Code / Aider

gmux-system's README lists "OpenCode, Claude Code, Aider" as supported agents. The integration is straightforward for any agent that exposes a similar HTTP API. The status detection for agents without an API falls back to pattern matching on terminal output — less precise but functional.

Layer 3 · interaction tools

This is gmux's lane, and it's comparatively empty.

DeepSeek-TUI (25k ⭐, +21,752/week in May 2026)

The most viral terminal AI tool right now. A Rust TUI for running DeepSeek models in your terminal. Fast, clean, well-designed. It's a single-agent terminal interface — one model, one conversation, one pane. There's no multi-agent concept, no gesture control, no live state aggregation. Its viral growth shows the appetite for terminal-native AI tooling is real.

Relationship to gmux: Not a competitor. DeepSeek-TUI would run inside a gmux-managed pane, and gmux would show its state (working/waiting) alongside all your other agents.

Warp (15k ⭐)

Warp is a commercial AI terminal with a polished UI, inline AI suggestions, and collaborative features. It's for developers who want a better terminal experience with AI built in. It's a single-session tool — no concept of managing multiple AI agents simultaneously. Cloud-dependent.

Relationship to gmux: Different philosophy. Warp replaces the terminal with a more polished experience. gmux wraps the terminal with a control layer. Warp is for individuals wanting a better single-agent experience; gmux is for multi-agent workflows.

Zellij (21k ⭐)

Zellij is a tmux alternative with a better UI, floating panes, and a plugin system. No AI integration. Its plugin architecture could theoretically be used to build something like gmux's status bar, but nobody has.

Relationship to gmux: gmux could theoretically wrap Zellij instead of tmux. No work has been done on this. tmux is the implementation for now.

tmux (37k ⭐)

The base. gmux wraps it. tmux handles session persistence, window/pane management, and the PTY infrastructure. gmux adds the AI awareness layer on top without modifying tmux itself — the status bar is injected via tmux's own status-right and window-status-format configuration.

The trending signal

In May 2026, the .claude/skills/ category exploded. Two repos — mattpocock/skills (+13k stars in one week) and addyosmani/agent-skills (+12k stars in one week) — both about packaging agent capabilities as open-source skill libraries, trended simultaneously. The community is actively defining what it means to package and share agent capabilities.

gmux's SKILL.md convention (used in graphify, compatible with Hermes) is directly in this space. A fivelidz/agent-skills repo publishing the best gmux-ecosystem skills (knowledge graph queries, memory injection, agent routing, gesture vocabulary definitions) would land in a trending category at exactly the right moment.

The honest competitive position

gmux's combination — gesture control + voice routing + live AI state across multiple agents + phone remote — is not replicated by anything in the current ecosystem. That's partly because multi-agent interaction at this scale is still new; most developers haven't hit the 10-agent problem yet.

The risk isn't competition from similar tools. The risk is that the tooling improves fast enough that qalcode2 (or whatever succeeds it) grows a multi-agent workspace mode natively, absorbing the status bar and session management features and leaving only gesture/voice as gmux-specific. The gesture and voice angle is hard to absorb into a coding agent tool because it requires camera/mic access and a separate visual layer. That's gmux's durable differentiation.

The window to establish gmux as the standard multi-agent terminal layer is open and currently uncontested. DeepSeek-TUI's viral growth proves the audience exists. The Python stack is already shippable. The Tauri app is close. The main blocker is publication.

Summary table

gmux: pip install gmux | gmux.ai