Learn Empryo in the terminal
Interactive demos of the real terminal app. Press the keys yourself, or let them play.
When a lesson pauses, press the key it shows or click the prompt. Type /onboard in the app to come back here. Not installed yet? Set it up first.
Ask about your codeLink to this section
Type a question and press Enter. Empryo finds the code through the Genome, reads only the lines it needs and answers with the files and symbols it used. Each step appears in the tool rail with its time, and the turn ends with its cost.
Find any commandLink to this section
Ctrl+K opens the command palette: every command, panel and setting in one searchable list. Type a few letters to filter it. / in the prompt does the same for slash commands.
Sign in with your subscriptionsLink to this section
/login lists every account Empryo can use: Claude Pro and Max, ChatGPT, GitHub Copilot, Grok, Gemini and more, with the ones already signed in marked. Pick one, approve it in your browser, and its models join the model picker. API keys go in /keys.
Switch models and effortLink to this section
Ctrl+L opens the model picker for every provider you have connected, subscriptions included. Type to search, Enter to switch. /effort sets how hard the model thinks. Each tab keeps its own choice.
Tune each providerLink to this section
/provider-settings holds each provider's own options: thinking, effort, how long the prompt cache lives and provider-side tools like web search. Tab moves between the providers and their options, Enter changes a value, and each change saves to this project or to every project.
Route each job to a modelLink to this section
/router gives every job its own model: the chat, explorers, coders, code review, web research and prompt preparation. Set how many subagents run at once, up to unlimited, and add backup models for when a provider fails.
Keep going when a provider failsLink to this section
Model Fallback, at the bottom of /router, gives each model a list of backups. When a provider is overloaded or rate limited, Empryo retries the same model, then switches to the next backup and finishes the turn on it. /retry sets how many retries come first.
Run subagents in parallelLink to this section
Ask for parallel work, or let Empryo decide. It splits the task, runs subagents side by side, each on the model its job needs, and merges what they report. Findings one subagent makes reach the others at their next step.
Connect agents in a workflowLink to this section
/orchestrate runs a graph of agents: a planner, a coder, two reviewers in parallel and an arbiter that sends failed work back for another round. Every step can use a different model. /runs follows the graph live.
Ask the GenomeLink to this section
The Genome is a live map of your code: symbols, calls, imports and git history. Empryo knows what depends on a file before it edits it, and tells you. /genome shows what is indexed.
Map the request firstLink to this section
The Marionette reads your request before the agent starts and turns it into a map of the repo: the files and symbols it touches, how they connect, and what it could not tell. The agent starts from the map instead of searching. /marionette <request> maps one request, and /marionette on maps every one.
Bring your own agentsLink to this section
/agents lists the agents defined as Markdown files in your project and your home folder, each with its own instructions, model, mode and tools. Empryo reads their descriptions and hands them the work that fits: in a dispatch, in a workflow or on their own.
Plan before it editsLink to this section
Ctrl+D cycles the modes. In plan mode Empryo researches and writes a plan without touching a file. Read the steps, then implement, revise, or hand the plan to a workflow.
See every change and go backLink to this section
/changes lists every file the agent touched with its diff. Checkpoints hold both the files and the conversation, so you can step back to any earlier turn.
Steer while it worksLink to this section
Type while the agent runs and press Enter: your note reaches it at its next step, without stopping the turn. Ask it something and it answers right under your note, then carries on. Alt+Q queues a follow-up instead, and Ctrl+X stops the turn.
Get a fresh reviewLink to this section
/review starts a separate reviewer on what this tab changed. It reads the diff without the conversation's assumptions and reports findings with the file and line.
Talk it through with the duckLink to this section
/duck opens a rubber duck in a new tab: a side chat that cannot see your code, for thinking out loud about names, trade-offs and plans. /duck <question> asks it straight away. /advise gets a second opinion on the whole conversation, and /council lets two to four models debate a hard call.
Work in several tabsLink to this section
/tab new (or Ctrl+T) opens a tab with its own conversation, model and mode on the same project, and /tabsplit shows two side by side. Ask one tab to check with another: it asks, and the other tab answers in its own pane.
Pick up where you left offLink to this section
Ctrl+P lists your saved sessions with their titles and size. Open one to continue exactly where it stopped. /memory keeps decisions and lessons between sessions.
Work from your phoneLink to this section
/remote turns on phone access and shows a QR code. Scan it with the Empryo app on your iPhone or iPad and the phone joins this session over Wi-Fi. A prompt you send from the phone runs here, with this project, its tools and your models, and the answer shows on both screens. /remote stop turns it off. The code and address in this demo are stand-ins, so the QR code does not scan; run /remote in your own terminal for a real one.
Let it use your appsLink to this section
Ask for something outside the terminal and Empryo drives the app itself. It lists what is open, reads the window's controls the way a screen reader does, then clicks and types through them by name, several steps at a time. /computer shows what it can reach and what your system still has to allow, and /computer stop takes the controls back.
Add skills, tools and routinesLink to this section
/skills loads playbooks for specific jobs, /mcp connects external tools, and /schedule runs a routine on a timer, like a morning check of your project.
Keep the prompt cache warmLink to this section
/cache shows how long your provider keeps the prompt cached, with a live countdown. /ping refreshes it while you are away, so the next turn stays cheap.