Empryo documentation

Start hereLink to this section

If you prefer to answer a few questions and get tailored commands, use the guided setup.

What you can doLink to this section

• Understand a project. Ask how a feature works and get answers that point to specific files. Empryo searches the Genome, reads symbols, and uses language servers where they are available.
• Make focused changes. Describe a fix or refactor. For TypeScript and JavaScript, Empryo edits functions and classes by name instead of matching raw text. Other languages use text edits plus whatever your language server supports.
• Check the result. Ask it to run your tests, linter, or type checker. Then review the output and the diff before you commit.
• Change Empryo's settings by asking. Ask it to switch models, turn on thinking, or connect an MCP server (a plugin that gives the agent extra tools). You choose whether the change applies everywhere or only to this project.
• Pick up where you left off. Sessions are saved so you can resume them. Checkpoints let you roll back agent changes, but keep using version control as your real backup.

Try this as your first prompt:

Explain the project entry points and cite the relevant files. Do not change anything.

More it can doLink to this section

• Work on several things at once. Each tab runs its own agent: up to 10 in the terminal and 32 in the desktop app, each in its own git worktree if you like. Steer a turn while it runs.
• Check its own work. The goal loop repeats until a separate reviewer passes the result, and a council of 2 to 4 agents can argue a hard decision.
• Reach past your code. It drives a browser, uses your other apps through computer use, and takes spoken prompts.
• Run on a schedule. Routines run a prompt on a timer, such as a morning check of your project.
• Fit it to your work. Morphs and cells add panels, boards and rules. The Marionette maps the files that matter before each prompt, and TypeSafe makes small, quick decisions with a small model.

Three ways to workLink to this section

All three use the same agent, settings, and sessions. Their controls differ. See installation for supported platforms and headless workflows for automation.

Add more when you need itLink to this section

Task routing (a different model per type of task), parallel agents, skills, and MCP servers are all optional. Start with one model and a small task, then add what your work needs.

Need help? Troubleshooting · FAQ · GitHub issues.