Tutorials¶

Hands-on, end-to-end walkthroughs — one per demo. Each is self-contained and copy-pasteable; outputs shown are from live runs on the KVM node behind https://beta.barista.sh.

Tutorial	What you run
1 · Counter	A number that survives pause→resume (memory, not disk) — the "aha"
2 · Claude Code session	Claude Code inside a session; attach, pause, resume
3 · Codex session	Codex inside a session; attach, pause, resume
4 · Sessions via MCP	Drive the session lifecycle from an MCP client (e.g. Claude Code)
5 · Session in Zed (ACP)	Drive a session's agent from Zed over the Agent Client Protocol
6 · Agent in a pausable sandbox	End-to-end: agent codes in a sandbox → pause/resume → `barista sync` the result down
7 · Templates	Build once, register a name, create by name — no digest hunting
8 · Python SDK	The lifecycle from Python (typed errors, wake-on-exec) + the E2B migration path
9 · Private images	Fleet registry allowlist, the 422, and the one operator step for private pulls

One-time setup (used by every tutorial)¶

You need the barista CLI, an API key, and (for create) a running Docker so the CLI can resolve image digests.

# from the repo, until it's on a package index:
alias barista='uv run barista'      # or: python -m barista_cloud.cli

export BARISTA_URL=https://beta.barista.sh
barista login <your-api-key>              # saved to ~/.config/barista/key
barista ls                                # → {"tenant":"you","sessions":[]}

Get a key from the console (/app/keys) or an operator (demos/seed.py).

The console's API keys page — create a key here; the secret is shown once at creation

Every session you create in these tutorials also shows up live in the console (/app), with pause/resume, exec, and the event stream one click away:

The console's Sessions page with a running session

Background: the engine's own concepts (sessions, sleep-and-wake, snapshots) live on the barista.sh docs site; the full CLI reference is ../using-barista-cloud.md.