Supported agents

Mastery HQ drives thirteen agent types. Eight are CLI agents that run locally with your own accounts; OpenRouter, Qwen, Kimi and Local models are HTTP backends (any model, no CLI); Hermes is the built-in orchestrator. Settings → Agents shows live detection of what’s installed.

Agent	Runs via	Notes
Claude Code	claude CLI	The default agent brand
Codex	codex CLI	OpenAI’s coding agent
Cursor Agent	cursor-agent CLI	No native Windows build yet — runs via WSL (Settings → Agents links the docs)
Gemini CLI	gemini CLI	Google; multimodal models
Aider	aider CLI	Pair-programmer; one-click install (the app fetches a private Python via uv — no admin, nothing touches your system)
GitHub Copilot CLI	copilot CLI
Grok Build CLI	grok CLI	xAI’s coding agent; signs in with your xAI key (XAI_API_KEY) or its own grok login
OpenCode	opencode CLI	Open-source agent; a model picker in the window header switches its model per window (opencode run -m)
OpenRouter	HTTP (no CLI)	Any OpenRouter model; needs OPENROUTER_API_KEY
Qwen	HTTP (no CLI)	Alibaba’s Qwen models over OpenRouter — Qwen3.7 Max/Plus/Flash plus the coder and thinking variants; needs OPENROUTER_API_KEY
Kimi	HTTP (no CLI)	Moonshot AI’s Kimi models over OpenRouter — K3, K2.7 Code, K2.6, K2.5, K2 Thinking; needs OPENROUTER_API_KEY
Local model	HTTP (no CLI)	Ollama, LM Studio, Jan (auto-detected, no key) or any OpenAI-compatible endpoint (Msty, Together, Groq, a gateway). See local & custom models
Hermes	Built-in	The orchestrator — dispatches to other agents and drives the workspace

Keeping the CLIs up to date

Settings → Agents shows the published version beside each installed CLI and offers to bring it up to date. It checks when you open it and every few hours.

It never updates anything on its own. Mastery HQ drives these CLIs with specific flags, so an upgrade arriving unannounced could break agent launching with nothing obvious to connect it to — you decide when.

What the row offers depends on how the CLI got there:

• Mastery HQ installed it, or the CLI can update itself (Claude Code, Codex, Grok Build) — Update does it, and the log streams as it goes.
• Something else installed it — a system npm install -g, or the vendor’s own installer — you get the exact command to run instead of a button. A button there would run, report success, and change nothing, because Mastery HQ can only write to its own install area.

Whichever route it takes, the CLI is run once before the update is called done, so an install that finishes “successfully” but leaves a command that won’t start is reported as the failure it is.

Model quick-picks stay current

The model dropdown in an agent window offers that brand’s current models. New models are folded in automatically as they’re published, so the list doesn’t quietly go stale between releases:

• · new — genuinely newer than anything that shipped with the app for that model line.
• · also available — a model line the built-in list never covered. It exists; that isn’t the same as it being newer.

The built-in picks stay first and in order — a few models are deliberately left out because they’re rejected on a normal sign-in — and Custom… still takes any model id your account supports. If the catalog can’t be reached, you get exactly the built-in list.

How sign-in works

CLI agents use their own authentication (whatever you’ve signed into in that CLI) or an API key from Mastery HQ’s local .env — keys are forwarded to the CLI at launch and never leave your machine. Each brand’s window carries its own accent color so you can tell a Claude from a Codex across the canvas at a glance.

Picking a default

Settings → Agents → Default agent sets which brand “open an agent” spawns.