How Maylyn works
The command reference and the core concepts. The deep guides live in the repo and in the app, so they never drift from the code.
Command reference
Run maylyn help, or maylyn <cmd> --help for a command's own flags.
Connectors, skills & slash commands
Three ways to add capability. They compose — a skill can tell the model when to reach for a connector's tools.
MCP connectors
Maylyn speaks the Model Context Protocol, so any MCP server works. Tool calls still pass the approval gate and run sandboxed.
Skills
A SKILL.md of task instructions the model loads only when a task needs it. Follows the Agent Skills spec.
Slash commands
Built-in verbs that mean the same thing in the terminal and the web. maylyn slash lists them all.
Built-in connectors
Also in the catalog when their host tool is present: playwright, chrome-devtools, database, docker, media (yt-dlp), github, context7, memory, filesystem, sequential-thinking. The Library's MCP tab also offers youtube-transcript, slack, notion, google-maps, email and yahoo-finance; those that need credentials ask for them at install.
Add an MCP server
The sidecar only learns tool names and schemas — every call RPCs back into Maylyn's Go registry, so sandbox containment, the safety kernel, and the approval prompt are identical to a built-in tool.
Providers
Local by default. Any OpenAI-compatible server works; add a hosted key any time.
MAYLYN.md
Drop a file in a project directory; Maylyn loads it every session started from there.
Two files stack: this per-project MAYLYN.md (standing rules, read from the workspace root) and a NOTES.md scratchpad the model can append to.
Talk to it
Local text-to-speech and speech-to-text. Enable it at setup.