EmpirioLabs AI CLI

empirio starts your coding agent already connected to EmpirioLabs AI. One command signs you in, picks a model the agent can run and starts the agent. Your agents’ own settings stay as they are, and the model list comes from the live catalog, so new models are available the day they launch.

It also runs any EmpirioLabs AI tool straight from your terminal, such as image, video and speech generation, web search and document reading: see Run tools from the terminal.

Install

curl -fsSL https://empiriolabs.ai/install.sh | sh

The installers check every download against the release’s signed checksums before installing anything. You can also download an archive for your platform from GitHub Releases, or build it with Go: go install github.com/EmpirioLabs-AI/cli/cmd/empirio@latest.

Start an agent

empirio

empirio on its own opens a menu: choose an agent, choose a model, and the agent starts. To start one directly, name it:

empirio claude                          # Claude Code
empirio codex                           # Codex CLI
empirio gemini -m <model-id>            # Gemini CLI on a model you choose
empirio claude -p "explain this repo"   # arguments after the name go to the agent
empirio codex exec "add tests"

The first run opens your browser to sign in. After that, empirio starts the agent straight away.

Supported agents

Install an agent with its own installer, then run its empirio command. empirio doctor lists the agents installed on your computer.

Sign in

empirio login

Your browser opens the EmpirioLabs AI consent screen. Approve EmpirioLabs AI CLI, and the computer appears under Settings > Connected applications in the dashboard, where you can disconnect it at any time. empirio logout disconnects it too.

Create keys on the API keys page.

Models

Each agent starts on your default model, or on the catalog’s first featured model it can run when you have not chosen one. An agent only lists models that support what it needs: the API it speaks and tool calling.

empirio models                    # every model with its context window and prices
empirio models --agent codex      # the models Codex CLI can run
empirio model <model-id>          # make a model your default
empirio claude -m <model-id>      # one session on another model

Agents with a second model for quick background work, such as titles and summaries, get the lowest-cost featured model by the same author. Choose another with --fast-model <model-id>.

EmpirioLabs AI tools

Agents marked Yes under EmpirioLabs AI tools in the agent table also get the tools of the EmpirioLabs AI MCP server, with nothing to set up. Each session gets the toolsets marked Yes under In the default; a tool call is billed like the same API request.

empirio claude --tools all             # every toolset, including GPU Cloud and hosted agents, which bill while they run
empirio codex --tools search+media     # only these toolsets
empirio opencode --tools none          # no EmpirioLabs AI tools

Set EMPIRIO_TOOLS to choose for every session. The tools are added next to the agent’s own MCP servers for that session only; a server you already named empiriolabs is used instead. Each toolset’s tools are listed on the MCP server page.

Run tools from the terminal

empirio run runs any EmpirioLabs AI tool directly: generate images, video and speech, search the web, read documents, check jobs and more. The tools come from the EmpirioLabs AI MCP server, so new tools are available without updating empirio. It uses your sign-in (or EMPIRIOLABS_API_KEY), and each run is billed like the same API request.

empirio tools                          # every tool you can run, by toolset
empirio tools generate_image           # one tool's arguments
empirio run generate_image prompt="A lighthouse at dawn, watercolor"
empirio run web_search model=tavily-search query="open-weight model releases this week"
empirio run generate_video prompt="Waves rolling onto a beach at sunset" --out clips

• Arguments are name=value pairs, typed by the tool: numbers, true or false, and JSON for lists and objects. Pass a whole object with --json '{...}' or --stdin. empirio run <tool> --help shows what a tool takes.
• Files: generated images, video, audio and 3D models are saved in the current folder, or in --out <dir>, and never overwrite a file. --no-download prints the result only.
• Long jobs: a video or another long job is followed until it finishes, for up to 10 minutes. Change that with --wait <seconds>; --wait 0 returns the job ID at once, and empirio run get_job job_id=<id> gets the result later.
• Output: the tool’s result is printed; --json-output prints the whole result as JSON.

Commands

Your agent settings

empirio passes the connection to an agent for that session only. Where an agent saves a choice of its own during a session, such as a model picked with /model, empirio puts back the value it found when the session ends. empirio <agent> --print-env shows exactly what an agent receives.

empirio keeps its own settings and your sign-in in %APPDATA%\empirio on Windows and ~/.config/empirio elsewhere, readable only by you.

Update and uninstall

empirio update installs the latest release after checking its signature and checksums, and the CLI tells you once a day when a new release is out. empirio uninstall removes the program and its settings and disconnects a browser sign-in. Agents and their own settings are left as they are.

Verify a release

Every release publishes SHA256SUMS, its signature SHA256SUMS.sig, and a build provenance attestation for every file. With allowed_signers from the CLI’s repository:

ssh-keygen -Y verify -f allowed_signers -I release@empiriolabs.ai -n empirio-release -s SHA256SUMS.sig < SHA256SUMS
sha256sum --check --ignore-missing SHA256SUMS
gh attestation verify empirio_linux_amd64.tar.gz --repo EmpirioLabs-AI/cli

Troubleshooting

Building an app that signs people in to EmpirioLabs AI? See Sign in with EmpirioLabs AI.