Quickstart

Go from install to your first reviewed change in about five minutes: connect a model, open a project, ask a question, and check the diff.

Get started4 minutes to read

Copy & share

On this page7 sections

This page gets you from a fresh install to a first reviewed change. Install Empryo first, and use a project you own or trust.

1. Connect a modelLink to this section

Pick one option. Run the commands below in your own terminal.

Desktop app: on first launch, a setup card asks you to pick a provider, enter a key or sign in, and choose a model. A default model is already selected. If Empryo finds a key you already have (for example in an environment variable), it skips this card. After that, a short tour shows you the terminal, browser, and codebase map, and then offers a few starter prompts. You can skip the tour.

Terminal UI: type /login to pick an account to sign in with, or /keys to manage API keys.

bash

empryo --login claude

For ChatGPT Plus, Pro, or Team, use empryo --login codex. If the browser can't reach Empryo on your machine (for example over SSH), use empryo --login codex-device to sign in with a code instead.

Other accounts you can sign in with include GitHub Copilot, Grok, Cursor, Devin, Google Gemini CLI, Google Antigravity, Kimi Code, OpenRouter, and LLM Gateway. Type /login to see the full list. Usage limits depend on your plan and provider.

For example, with an Anthropic API key:

bash

empryo --set-key anthropic YOUR_API_KEY
empryo --list-providers

Replace YOUR_API_KEY on your own machine. Never paste keys into a website or commit them to a repository. Your provider bills API usage separately from any consumer subscription. See providers for other key-based options.

Install and start Ollama, then download a model that fits your hardware:

bash

ollama pull llama3.3

Keep Ollama running and select its model in Empryo. LM Studio works too. Local models need no API key, but they use your own memory and compute.

2. Open a projectLink to this section

In a terminal, go to your project and start Empryo:

bash

cd your-project
empryo

In the desktop app, open the project folder.

The first time you open a project, Empryo asks whether you trust it. Read this prompt carefully. A trusted project's own config can add providers, MCP servers, and other settings that run code.

To pick or change the model in the terminal UI, press Ctrl+L. Empryo then builds the Genome, its map of your code. How long that takes depends on the size of the repository.

3. Ask a questionLink to this section

text

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

Read the answer and open the files it points to. A focused question about a real feature or a failing test is a better start than a broad rewrite.

4. Make a small change and review the diffLink to this section

Pick a real issue in your project and ask:

text

Propose the smallest fix for this issue. Explain the plan before editing, then run the relevant tests and show what changed. Do not commit.

Review the diff and the test output. Tests can fail, and some checks may not be set up in your project. Commit only when you're happy with the result.

5. Change settings by askingLink to this section

You don't need to find a config file. Ask in plain words:

text

Use adaptive thinking, and make the diffs compact for this project only.

Empryo checks the current value, writes the change, and tells you what changed and where. Settings for every project go in ~/.empryo/config.json. Settings for one project go in .empryo/config.json inside that project. Most changes apply right away. If a setting only takes effect after a restart, Empryo tells you.

You can also ask about Empryo itself:

text

Why does my context keep filling up, and what would you change about my setup?

Signing in and adding API keys stay manual on purpose. Use /login, /keys, or empryo --set-key.

Headless alternativeLink to this section

To run a single prompt from a script, go to your project directory after connecting a model and run:

bash

empryo --headless "Explain the project entry points without changing files" --json

Run empryo --list-models to see model IDs, and add --model provider/model to choose one. In automation, --trust marks the repository as trusted permanently, so only use it on checkouts you control. See headless workflows.

Next stepsLink to this section

Type /help to see commands and shortcuts, or ask the agent what a command does.
Try task routing if you want different models for different kinds of work.
Add MCP servers or skills when a task needs them.
Learn about checkpoints, and keep using version control as your backup.