Developer workflow

Development against any is one long-running server process on your machine and whatever clients you point at it. This page is the loop you will repeat: init, run, call, inspect.

1. Init an account

any init

Creates ~/.any/ (mode 0700), generates an account, writes ~/.any/<accountId>/wallet.key, and prints the BIP-39 mnemonic to stderr once. Back it up. Re-running any init when an account exists is a no-op that lists them.

To attach a second machine to the same account, restore from the phrase — the account id is the same, the device key is fresh:

any init --mnemonic-stdin < phrase.txt      # never copy wallet.key between machines

2. Run the server

any run                                     # foreground, 127.0.0.1:7001
any run --addr 127.0.0.1:0                  # ephemeral port; prints "LISTENING <addr>" on stdout
any run --account <id>                      # pick one when the data dir holds several
any run --mode managed                      # a host-owned server: phrase per launch, control token

any run is foreground-only: run it in a terminal, tmux, or a user service. It never creates a wallet — on a fresh data dir it starts unauthorized and every route except /v1/health, /v1/shutdown, /v1/openapi.json, /v1/auth answers 401 auth.required until POST /v1/auth (or any auth login) boots an account in place. The default standalone mode is the one to develop against; managed is how an app shell embeds the server (Accounts).

Stop it with Ctrl-C or:

any stop                                    # finds the server by its account lock, SIGTERM —
                                            # drains in-flight, 10 s deadline

3. Check health

any status                                  # GET /v1/health

{ "status": "ok", "version": "any v0.1.2 (commit 1a2b3c4, built 2026-09-09)",
  "startedAt": "2026-08-24T09:00:00Z", "networkId": "N83gJpVd…",
  "account": "A8tR…", "bootstrapping": false }

networkId is the any-sync network the server joined (Networks). bootstrapping: true means the background space-loading pass is still running; the server is serving already. crdtVersion ({supported, stored, newer}) is the account's data-model version: every release stamps its own on first open and the mark only rises, so a release older than the data refuses to boot (409 sdk.crdt_version_newer) or, when the raise arrives from another device while running, turns read-only — newer: true is the client's "upgrade required" signal.

4. Call it

Everything is http://127.0.0.1:7001/v1/… with JSON bodies. Pick whichever client fits:

any --verbose space get $SPACE              # log the HTTP exchange to stderr
any --addr 127.0.0.1:7002 status            # a second server on another port

Without --addr the CLI finds the server serving the data dir's account through its server.addr file, so an ephemeral-port server needs no flag. When the server is down the CLI exits 3 with start it with any run in another terminal; it never auto-starts one.

5. Inspect

• Sync — any sync-status space $SPACE, any sync-status subscribe.
• Search index — any search $SPACE "reranker"; check vectorStatus in the reply.
• Raw datasets — any query-subscribe $SPACE $OBJ --dataset chat_messages --limit 20 --sort='-_ver.id' prints one JSON line per SSE frame.
• Diagnostics — any debug space $SPACE / any debug object $SPACE $OBJ dump head-sync counters and per-object tree state (unstable output, diagnostic only).
• Logs — one stream on stderr through the SDK logger; --log-level debug or ANY_LOG_LEVEL=debug. Add an absolute file path under log.outputPaths in config.yaml to tee to a file (Debugging).

The data dir

~/.any/                         # dataDir — a ROOT that can hold several accounts
├── config.yaml                 # optional
├── models/                     # shared embedder model cache (~600 MB, downloaded once)
└── <accountId>/
    ├── wallet.key              # standalone: account + device keys (0600)
    ├── device.key              # managed: this device's key only (0600)
    ├── server.lock             # single-instance lock (an OS file lock)
    ├── server.pid              # holder's pid, for error messages only
    ├── server.addr             # holder's bound address, read by the CLI
    ├── sdk/                    # any-store databases — owned by the SDK
    ├── files/                  # file bytes, one CARv2 per root CID
    └── index/                  # local search index — derived, safe to delete

Two servers may share a root as long as they serve different accounts on different ports. The index/ directory is derived state: removing it with the server stopped is safe, and the next start re-indexes every space from the change feed. Never delete files/ by hand — a file that has not been backed up yet has its only copy there (Data dir).

Config precedence is file → ANY_* env → flags; the file is looked up at --config, $XDG_CONFIG_HOME/any/config.yaml, ~/.config/any/config.yaml, then <data-dir>/config.yaml (Configuration).

Building from source

make build                                  # → bin/any (+ bin/llamacpp for the local embedder)
go test ./...

make build adds the local embedder (-tags llamacpp) and its llama.cpp libraries; a bare go build serves full-text search, and vector search only through an online provider named by index.openai.*. go test ./... runs everything except the local-embedder tests, which make test adds. On NixOS run builds and the binary through nix develop -c … so the local embedder finds libffi (Builds and CI).

Note. An unconfigured binary syncs against the production network. For experiments use a staging or local nodeconf via ANY_NETWORK_NODECONF_PATH, or the embedded placeholder that joins no network at all (Networks).