FAQ

Do I need my own Claude subscription?

Yes. Command Center is bring-your-own-Claude: the agentic layer runs through Claude Code (the claude CLI), installed and signed in under your own account. Your agents run on your subscription, your rate limits, and your own data relationship with the model provider — model calls go directly from your machine to the provider, with no relay in between. Command Center's subscription covers the product; it does not include model usage. See BYO Claude & keys for what runs on whose account.

Which API keys does Command Center need?

None to boot — the app, the graph, keyword memory search, documents, and terminals all work without any key. Two optional Gemini slots unlock extras, both set in Settings ▸ API Keys:

• GEMINI_API_KEY (free tier) — semantic memory search, knowledge-base embeddings, and Gemini queries. Full-depth workspaces need it at search time as well as at build time.
• GEMINI_IMAGE_API_KEY (paid tier) — image generation only. Without it, image generation is unavailable; the imagegen service itself still runs and reports itself unconfigured, so a dark image feature is not a dead daemon.

Is there a free tier or a trial?

No free tier. Command Center requires sign-in with an active subscription — Pro or Ultra — to unlock the product surface. See Tiers: Pro & Ultra.

What is the difference between Pro and Ultra?

Both are the same full local install; your tier decides which feature surfaces unlock after sign-in. The current split is documented on the tiers page — the docs keep that page as the single place the answer lives, so it cannot drift.

What runs on my machine, and what runs remotely?

Almost everything is local: the desktop app, all background daemons (bound to loopback 127.0.0.1 only), the ArcadeDB knowledge graph, your documents, agents, and configuration. The connections the platform itself initiates are: agent runs calling your model provider on your own subscription, the API calls you configured with your own keys, the sources the intelligence desk fetches because you added them, and sign-in plus the update check talking to agenticonsult.de. Anything else that leaves the machine is something you configured or an agent you dispatched.

What happens when I go offline?

Local work continues — files, terminals, the graph, anything that does not need a provider. Your sign-in keeps the machine unlocked for up to 72 hours from its last successful verification; in practice that is 36 to 72 hours, because it renews at the halfway point rather than waiting for expiry. There is no extra grace period afterward — the tolerance is the sign-in's own lifetime. Update downloads keep working for a further four days on purpose, so security fixes still reach a machine whose features have paused.

Past that window, tier-gated views lock and Doctor's Account row reads degraded (offline_too_long). Reconnect and click Refresh and everything unlocks immediately. Nothing is deleted at any point; your files and local data are untouched.

How do I move to a new machine or reinstall without losing work?

Your state is the home folder. Copy it to the new machine (or keep it in place for a reinstall), install the app, and point it at that folder — at first run, or later via the project root in Settings. The app lays down anything missing and preserves everything present. Afterward, open Settings ▸ Doctor and press Provision in the Runtime group once: Python environments are machine-specific and are rebuilt rather than copied.

One thing to plan for: one subscription covers two machines. Signing in on a third is allowed and does not fail — instead, the least recently used of your existing machines is signed out server-side and steps down to the signed-out baseline at its next refresh (its data stays fully accessible). Reinstalling does not free a slot; a machine is released by another machine taking its place. Details on requirements.

Where do my files live?

In one home folder you choose at first run. Inside it: documents/ (your content — reports, missions, notes), config/ (runtime configuration and version stamps), the app folder's .env (your secrets, local only), each component's data/ directory (local databases), and the engines' own folders. Desktop preferences (window layout, theme) live in your operating system's per-user app-config directory, and the auto-memory layer lives beside your agent CLI's own per-project state rather than inside the home folder. There is no cloud copy of any of it.

Will an update overwrite my customizations?

No. Updates overlay the shipped code files and seed new agents, skills, and protocols — but a doctrine file you edited is never overwritten, because an agent (or you) tuning its own doctrine is the product working as designed. Your documents/, .env, config/ state, databases, Python environments, and engine installs are structurally out of the update's reach. Restoring the shipped doctrine is always an explicit action (Settings ▸ Doctor ▸ Refresh doctrine) that backs up your copies first. Details: How updates work.

How do I reset the app but keep my content?

The reset levers are separable, so you never have to nuke everything:

• Doctrine — Settings ▸ Doctor ▸ Refresh doctrine restores the shipped agents/skills/protocols (your edited copies are backed up first).
• Platform tree — Doctor ▸ Re-run lay-down repairs a damaged or partial install in place.
• Python runtimes — Doctor ▸ Runtime ▸ Re-provision, or Repair on a single environment. The command-line equivalent (python scripts/provision_runtime.py) exists for when the app will not open.
• Knowledge graph — the graph is derived from your files and can be re-synced and re-ingested at any time.

documents/ is never part of any reset path.

Why are the video engines not installed by default?

Remotion and Hyperframes are opt-in, per-engine installs from the Video tab. Each carries its own license acceptance at install time and a heavyweight footprint (a Node dependency tree plus a render browser fetched on first render). The rest of the product does not depend on them, so they stay off your disk until you want video work.

Do I need Docker?

No. The default knowledge-graph backend is a provisioned-local ArcadeDB — a bundled Java runtime plus a pinned package that Command Center installs and manages itself (Settings ▸ Database). A docker/remote mode exists only if you prefer to point at an ArcadeDB server you run elsewhere.

Doctor shows several daemons as "not running" — is my install broken?

Usually not. "Not running" is an info state, not a failure: a daemon you are not using can stay off. Autostart is on by default (Dashboard ▸ Platform Health), so an all-stopped board usually means it was turned off for those services. Something is actually wrong only when a row is a red fail, or when a daemon refuses to start after you press Start — in that case see Troubleshooting.

I set up everything and nothing is being harvested or published.

That is the shipped state, not a fault: no routine is scheduled on a fresh install. Scheduling is per-routine state on your machine and it ships off, so the intelligence pipeline stays silent until you say what should run and how often. Open Routines, enable the ones you want, and expect roughly a minute of warm-up before the first run. Check the routines daemon is running while you are there — nothing fires without it. See Routines.

My own Send button stopped working. Drafts still save.

Check Settings ▸ Terminals ▸ Agent control planes ▸ Email sending. That switch is a master outbound stop for the machine: your composer does not go through the agent tool surface, but it does go through the same mail client underneath, and the stop lives inside that client. While it is off, nothing leaves by mail — agent or human — and reading, searching, and saving drafts keep working, so a message you compose waits as a draft instead of vanishing. It is deliberate: a stop with an exception is a stop that gets routed around. See Autonomy and controls.

Cortex opens but the graph area is a message instead of a canvas.

The graph canvas needs working GPU acceleration, and Command Center tests for it before drawing rather than crashing the view. The panel you are seeing names the reason. The usual causes are a virtual machine or virtual desktop, an RDP or screen-shared session, hardware acceleration disabled by policy, or a very old graphics driver.

Nothing is wrong with your data. The knowledge graph is unaffected — Overview, Report, Lint and Log read the same content without a GPU, ingest and search are untouched, and agents query the graph directly. There is deliberately no low-fidelity fallback drawing, because a misleading picture of your graph would be worse than an honest message. The fix is a host with acceleration: enable it where policy allows, connect locally rather than through a remote session, or use a machine with a GPU.

A tool or a view says "subscription required" even though I am signed in.

Two different things produce that, and Doctor's Account row tells you which. If the row is degraded, verification has lapsed — reconnect and Refresh, and see the sign-in cluster for what each reason means. If the account is healthy, the feature is simply above your tier: the refusal names the capability and your current plan and points at the upgrade page. It is a normal, designed answer rather than an error — agent tools return it as a result, not a crash. See Tiers: Pro & Ultra.

I saved an API key, but the feature still says the key is missing.

Daemons read keys at process start, so a key saved while a daemon is running is invisible to it until that daemon restarts (Dashboard ▸ Platform Health). For semantic memory search, restart Shared Memory there — or, if an agent session is currently hosting that surface, restart the session instead. Doctor detects this exact state and names it ("the running memory daemon has no key loaded").

My agent terminals stopped resuming after a restart. Why?

Session resume rides hook scripts installed into the agent harness, and missing hooks fail silently — nothing errors, resuming just stops happening. Check Settings ▸ Doctor for the "Fleet wake & session-resume hooks" row; if it reports hooks as not installed, run python .claude/hooks/cc_install_hooks.py from your install root. New terminal sessions pick the hooks up immediately.

Does Command Center upload my documents anywhere?

No. Your content leaves the machine only when you invoke something that needs it: a model call carries the context you gave it, under your own account and keys, directly to that provider. The subscription check exchanges account status, not content. Proxy credentials, if you configure a BYO proxy, stay in your local .env and are never transmitted anywhere except to your own proxy provider.