AI agents and skills
Coding agents, API keys in your keyring, local models, and skills for every agent.
On this page
AI agents and skills
Agents
BxCachy installs and launches coding agents (Claude Code, Codex, Gemini CLI, OpenCode) and orchestrators.
bxcachy ai list
bxcachy ai default claude-code --yes
bxcachy ai launch # the default agent, with its keys loaded
bxcachy ai shell # a pull-down terminal (Super+`)
bxcachy ai session start # a persistent tmux session
API keys
Keys are stored in your desktop keyring (BxCachy installs one if you have none), never in files or shell history.
bxcachy secrets set ANTHROPIC_API_KEY # typed hidden
bxcachy secrets list
bxcachy secrets run ANTHROPIC_API_KEY -- some-command
Each key is marked as an AI key or another kind, and agents declare which providers they accept (OpenCode can use several). Exporting keys globally is opt-in. Move keys between machines with bxcachy secrets export (encrypted with age).
Local models
bxcachy ai models list|ps|pull MODEL|rm|enable drives Ollama. bxcachy gpu doctor shows what your GPU needs (ollama-cuda or ollama-rocm).
Skills for every agent
Skills are small SKILL.md files that teach an agent how to do one thing well. BxCachy stores them once in ~/.agents/skills and links each one into the skills folder of every agent you have (Claude Code, Codex, Gemini CLI, OpenCode). A folder of your own with the same name is never replaced.
bxcachy skills install bxcachy boxlang matchbox --yes
bxcachy skills status
bxcachy skills search matchbox # the skills.boxlang.io hub
bxcachy skills install ortus-boxlang/skills/boxlang-developer~language-fundamentals --yes
bxcachy skills update --yes
| Name | What it is |
|---|---|
bxcachy | Five skills about BxCachy itself, shipped inside the program |
boxlang | BoxLang language fundamentals, from the hub |
matchbox | The MatchBox runtime, from the hub |
The hub (skills.boxlang.io) audits every skill. A skill that failed its audit is refused. Skills are optional in the wizard (AI page).