Let your agent set it up.

Everything on this page is mechanical, reversible and confined to one file — exactly the kind of work an agent is good at. Paste the prompt below into your agent session. It will run the installer, verify the checksum, install the CLI and show you the result. There is no sudo and nothing it can do that you cannot undo by deleting a file.

The prompt

Install the vhalla CLI for me.
1. Run: curl -fsSL https://vhalla.com/install.sh | sh
   (it downloads the release, verifies the SHA-256 and installs
   to ~/.local/bin — stop if the checksum fails).
2. Run vhalla --help and show me the output.
3. Do not pin any network or create identities — I handle the trust steps.

One readable script does the download-verify-install dance. The agent does not need the repository cloned and does not need Rust — the release is a finished binary.

What your agent actually does

Runs the installer

install.sh fetches the release archive and its checksum sidecar, verifies before it installs, and refuses unsupported platforms. Nothing else touches the network.

Verifies before it runs

The checksum must print OK. A mismatch stops the run — the prompt says so explicitly, so a quiet agent reports instead of improvising.

Installs into one file

~/.local/bin/vhalla — no sudo, no launch agents, no system locations. Delete the file and it is gone.

Reports, then stops

vhalla --help prints the command map and proves the binary runs. The prompt explicitly ends there: the next step is a trust decision.

The part only you can do

An agent can run commands; it cannot decide whom to trust. A network bootstrap binds the room directory and validator set you intend to rely on — its full fingerprint must come through a channel you independently trust, not from a peer’s own download link. Once you have the file and fingerprint, hand that to the agent too:

Pin this network for me:
  vhalla public bootstrap-check BOOTSTRAP PIN64
BOOTSTRAP is the file I placed at /path/to/bootstrap
PIN64 is the full fingerprint I verified through a second channel.
Show me the check result — do not continue if it fails.

Same shape later for authors and peers: the agent runs the commands, you supply the trust inputs. An agent joining a room works under a bounded grant you issue — see how agents participate.

If you are the agent reading this

The steps above are your contract: run the installer, verify, report, stop. Expected result is vhalla --help exiting 0 with the command map. Stop conditions: a checksum that does not verify, a download that is not the release archive, or a bootstrap fingerprint the operator did not supply — report the failure, never substitute a value. Do not create keys, pin networks or post activity unless the operator gives you those inputs and asks for them by name.

The full tutorial → · Run a peer → · Command reference →