Open-source AI infrastructure

Your AI runtime, from the terminal.

Install evs, connect to your Everstack instance, and take agent projects from local files to a running deployment. Work with isolated sandboxes from the same terminal.

Start with $5 credit Self-host Everstack

One-time starter credit per organization on Everstack Cloud.

Start with the binary.

One CLI for your local shell and remote Everstack instances. Choose the published release, then check the version before you begin. Before you start: macOS or Linux on ARM64 or x86-64, with bash and curl. The installer verifies the release checksum and may request sudo for /usr/local/bin. The CLI does not provision a backend or its dependencies.

Install the verified release

Uses the installer from the same release tag. Review the script before running it, or download a platform binary from the release page.

curl -fsSL https://raw.githubusercontent.com/everstacklabs/everstack/v0.1.25/build/install.sh | bash -s -- --version v0.1.25

Check your installation

These examples were checked against v0.1.25. If an older version appears, check for another evs earlier on your PATH.

evs --version

Point it at your instance.

Authenticate once, then reuse your active login for agent and sandbox commands. Connect to Everstack Cloud or your self-hosted deployment. Before you start: A reachable Everstack instance and an account with access. Replace the example hostname with your instance's API base URL, not the marketing site or dashboard URL.

Set your API endpoint

Replace this placeholder before running. Keep this shell open for the commands below.

export EVS_API_URL="https://your-everstack-host.example.com"

Sign in through your browser

Uses browser-based OAuth with PKCE. For a headless terminal, use the device-login command below.

evs login --api-url "$EVS_API_URL"

Confirm your active identity

Check that the identity and active context match the instance you intend to use.

evs whoami

Local files. Deployed agents.

Keep the agent definition, instructions, skills, and project functions in your repository. Deploy the directory, then stream a turn from your terminal. Before you start: Complete Connect first. Your instance needs an enabled agent runtime and access to the model configured in agent.yaml. Running agents can incur model and runtime usage charges.

Scaffold a project

Creates release-helper/ with agent.yaml, instructions.md, an example function, and an example skill. Choose a new directory name if this one already exists.

evs init release-helper

Review, then deploy

First edit agent.yaml to select an available model and review the generated instructions and function. This deploys to your active instance. Conflicting existing state is not silently overwritten.

evs deploy ./release-helper

Run a turn

Uses the agent name from agent.yaml and streams the reply. The input is positional, not a --message flag.

evs agents run release-helper "What can you help me with?"

A shell inside the boundary.

Create an isolated environment, execute a command, and follow its logs. Your shell sends requests to the sandbox backend configured on your instance. Before you start: Complete Connect first. Your instance needs a configured sandbox backend, a usable default image or template, and permission to create sandboxes. API-key sessions also need EVS_TENANT_ID for execution. Sandbox usage and retention may be billable.

Create an environment

Wait for the sandbox to be running. Copy its returned ID for the next step. Add --template with a configured template if your instance has no default.

evs sandbox create --name cli-demo

Run a command inside it

Replace the placeholder ID. The -- separator sends the remaining arguments to the sandbox command.

export SANDBOX_ID="replace-with-your-sandbox-id"
evs sandbox exec "$SANDBOX_ID" -- pwd

Follow sandbox logs

Streams sandbox logs, not gateway request traces. Press Ctrl+C to stop following; this does not stop or destroy the sandbox. Manage its lifecycle in the dashboard or CLI.

evs sandbox logs "$SANDBOX_ID" --follow

No browser on this machine?

Set EVS_API_URL to your instance, then request a device code. Complete authorization in a browser on another device if your deployment supports device login.

evs login --device --api-url "$EVS_API_URL"

Build it into your scripts.

Provide EVS_API_URL and EVS_API_KEY through your CI secret store. Sandbox list supports JSON output. Keep credentials out of command history and version control.

evs sandbox list --json

Verified command syntax

Checked against the published v0.1.25 binary on 2026-09-05. Feature availability depends on your instance, permissions, and edition.