Quickstart

Networks

A server syncs against exactly one any-sync network: a coordinator, tree nodes, file nodes, and their addresses, described by a nodeconf YAML. With nothing configured, the binary joins production. Decide this before the first any run of an experiment.

The nodeconf

networkId: N9DU6hLkTAbvcpji3TCKPPd3UQWKGyzUxGmgJEyvhByqAjfD
nodes:
  - peerId: 12D3Koo…
    addresses: [ "node1.example.net:6666", "quic://node1.example.net:8888" ]
    types: [ tree ]
  - peerId: 12D3Koo…
    addresses: [ "file1.example.net:6666" ]
    types: [ file, fileV2 ]
  - peerId: 12D3Koo…
    addresses: [ "coord1.example.net:6666" ]
    types: [ coordinator ]

Node type	Role
`coordinator`	Space registry, ACL/invite coordination, deletion, the 1-1 inbox
`tree`	Stores and relays encrypted object-tree changes; the peers head-sync talks to
`file` / `fileV2`	File byte storage. Durable file backup needs `fileV2` nodes; without them attaches stay `inflight`
`pubsubrelay`	Account/space-scope event bus relay

The networkId is part of every space's identity — a space created on one network cannot be moved to another.

Choosing one

Precedence is config file → env → flag:

# config.yaml
network:
  nodeconfPath: /etc/any/nodeconf.yaml
  # or inline:
  # nodeconf: |
  #   networkId: …

ANY_NETWORK_NODECONF_PATH=/etc/any/staging.yaml any run

You want	Do
Production — real data, real peers, other devices of your account	Configure nothing. The production nodeconf is compiled into the binary, so a packaged install boots from any working directory.
Staging — the same topology on throwaway infrastructure	Point `network.nodeconfPath` / `ANY_NETWORK_NODECONF_PATH` at the staging nodeconf.
Local infra — your own any-sync nodes on a LAN or in containers	Same knob, your own YAML. Everything works offline-first regardless, so partial deployments (no `fileV2`, no `pubsubrelay`) degrade feature by feature, never at boot.
No network at all — tests, demos, a single machine	Use the sanitized placeholder below.

Note. The production default is a convenience for installs, and a trap for experiments: a test script that forgets to set the env var creates real spaces on the real network under your real account. Set ANY_NETWORK_NODECONF_PATH in the shell you develop in, or put network.nodeconfPath in a per-experiment config.yaml passed with --config.

The placeholder: boots, serves, joins nothing

The repo ships internal/config/nodeconf-placeholder.yml: the real networkId with fake node addresses. A server started on it binds, serves every endpoint, and stores everything locally — but no peer is reachable, so nothing ever syncs and /sync-status stays offline. Tests use it (config.NodeconfPlaceholder()), and it is never selected at runtime by itself:

ANY_NETWORK_NODECONF_PATH=$PWD/internal/config/nodeconf-placeholder.yml \
ANY_DATA_DIR=/tmp/any-sandbox any run

Two devices on the same LAN still find each other over p2p discovery even on the placeholder — set p2p.enabled: false for full isolation.

Push pairs with the network

The push-notification node is a direct out-of-band peer, not part of the nodeconf. The packaged production push node is applied only when the network is also the production default; a staging or local server (and every test) gets no push node unless push.peerId / push.addrs name one explicitly, so nothing on a test network ever pushes through production (Push).

Embedded builds

any.aar and any.xcframework follow the same rule with one parameter: an empty nodeconfYAML in Start selects the embedded production conf (so a mobile host vendors no conf of its own and changes network with a binding bump); non-empty YAML text overrides it (Android, iOS).

Checking which network you are on

any sync-status space $SPACE        # networkPeers > 0 once a tree node is connected
any debug space $SPACE              # per-peer head-sync counters (diagnostic, unstable)

A server on the placeholder reports networkPeers: 0 and state: "offline" forever; a server on a real network moves to syncing / synced within a head-sync round.

Why it matters. Because every device holds the whole database, "which network" only decides who relays your ciphertext and to whom. The data model, the API, and the encryption are identical on production, staging, a LAN, or no network at all — which is what lets a test suite run the real server against a conf that goes nowhere.

See also: Networks (operations) for running your own nodes, and Configuration for the full network block.