Debugging

Most problems are one of four things: the server is not up or not authorized, a space has not converged yet, the index has not caught up, or a request was malformed. Each has a dedicated read.

$SPACE, $OBJECT and $CHAT below are ids from your own space; --addr points the CLI at a server on another loopback port.

Is the server up and authorized?

any status
curl -s http://127.0.0.1:7001/v1/health

{ "status": "ok", "version": "any v0.1.2 (commit 1a2b3c4, built 2026-09-09)",
  "startedAt": "2026-09-10T08:12:00Z", "networkId": "N83gJpVd…", "account": "A3…", "bootstrapping": false,
  "crdtVersion": { "supported": 2, "stored": 2, "newer": false } }

Add --verbose to any CLI command to see the HTTP exchange on stderr. Without --addr the CLI connects to the address the account's running server recorded in server.addr, so a server on an ephemeral port is found too; with several servers under one root, pass --addr.

Has this space converged?

/sync-status is the production read for sync state; it has a rollup per space, a per-object variant, and SSE streams (Sync status).

any sync-status space $SPACE
any sync-status object $SPACE $OBJECT
any space sync $SPACE          # force a head-sync round now instead of waiting ~30s

The debug endpoints go one level deeper and are diagnostic only — fields may move.

any debug space $SPACE                 # GET /v1/spaces/:spaceId/debug
any debug object $SPACE $OBJECT        # GET /v1/spaces/:spaceId/debug/objects/:objectId
curl -s http://127.0.0.1:7001/v1/debug/p2p

debug space returns per-peer outbound head-sync counters since boot (in-memory, reset on restart; peers is empty until one diff round has run):

{ "spaceId": "spc_…",
  "peers": [ { "peerId": "12D3Koo…", "lastSyncAt": "2026-05-15T12:00:00Z",
               "new": 2, "changed": 5, "lastErr": "" } ] }

debug object is a joint-consistent snapshot of one object's tree and sync state:

{ "objectId": "obj_…", "syncState": "syncing", "pending": ["head_a"],
  "lastSyncAt": "2026-05-15T12:00:00Z", "heads": ["head_a"], "headsCount": 1,
  "branchCount": 0, "treeLen": 3, "snapshots": 1,
  "latestVersionId": "01HX…", "maxAddSeq": 17 }

syncState is unknown / offline / syncing / synced / error. latestVersionId and maxAddSeq are local to this device and never comparable across peers.

Note. debug object locks the object tree and walks every change; on a very large tree it can pause local writes noticeably. Never poll it in a tight loop. First touch of a never-loaded object also triggers a cold restore.

/v1/debug/p2p shows the local-network layer — own peer id, listener, discovered LAN peers and which spaces they share (Networks).

Is search behind?

Long-running index work shows up in the process view:

any process list                       # GET /v1/processes

index.fts.<spaceId> is the change backlog (done counts changes, total unknown), index.embed.<spaceId> the vector drain with done/total, index.links_backfill.<spaceId> a rebuild of the link index, index.model_download the model fetch in bytes. Indexing announces itself only past three seconds of work, so ordinary edits never appear. On a search reply, vectorStatus: "unavailable" means the embedder is down or the model is still downloading; disabled means this server has none (Hybrid ranking). If boot fails with an index schema-version or dimension mismatch, remove <account-dir>/index/ and restart: every space then re-indexes from the beginning, visible as the processes above (How indexing works).

Reading an error

Every non-2xx response has one shape:

{ "error": { "code": "space.not_found", "message": "space spc_xyz not found",
             "details": { } } }

code is the machine-readable handle, grouped by area (request.*, auth.*, space.*, object.*, dataset.*, index.*, …); message is safe to show; details carries only identifiers the caller supplied — never secrets or paths. Common ones:

The full namespace is in the error reference. CLI exit codes: 0 success, 1 user error or 4xx, 2 5xx, 3 cannot reach the server.

No write sets a type or files a collection as a side effect — a value or dataset write is admitted only while the object's type or one of its collections declares it (Types and properties, Runtime datasets).

Logs

One log stream for the whole process, to stderr, level info by default.

any run --log-level debug
ANY_LOG_LEVEL=debug any run

log:
  defaultLevel: debug
  format: 2                  # 0 colorized | 1 plaintext | 2 json
  outputPaths: ["/var/log/any/server.log"]   # absolute; ~ is not expanded

5xx responses log at error with the full stack; panics are converted to 500 internal with a generic message and the trace goes to the log. A boot error saying index.embedder "local" needs -tags llamacpp means the binary has no local embedder — rebuild with make build, or use index.embedder: auto (Builds and CI).

Watching live

Every subscription is SSE, and the CLI prints one JSON object per frame, so the fastest way to see what a client sees is to attach the same stream:

any query-subscribe $SPACE $CHAT --dataset chat_messages --sort=-_ver.id --limit 20
any events subscribe --scope space --space $SPACE
any sync-status subscribe

Streams end with closed{reason} — server_shutdown, deauthorized, sdk_closed, overflow, drifted — and recovery is always "reconnect for a fresh snapshot"; after deauthorized, re-read GET /v1/auth first (Subscribe).

The embedded UI

A standalone any run serves a small debug harness at http://127.0.0.1:7001/ui (space picker, object tree, members, chat) — handy for a visual check of what the API holds. webUI.enabled: false turns it off; app-embedded boots run headless.