BYO Claude & keys
The product runs on your own Claude subscription and API keys — what that means in practice.
Command Center is the harness; the model is yours. Your subscription buys the platform — the shell, the orchestration, the memory fabric, the knowledge graph, the doctrine — and the intelligence that runs inside it comes from your own Claude account. That split is deliberate, and it is worth understanding exactly, because it is also the honest answer to "what will this cost me per month."
What runs on whose account
| Capability | Runs on | Notes |
|---|---|---|
| Agents, terminals, missions, routines | Your Claude account (a Claude subscription with Claude Code, or an Anthropic API key) | This is the platform's engine — every agent run is your Claude usage |
| The knowledge graph at structure and entities depth | No external account | Built mechanically or by your Claude agents; zero external API calls |
| Shared memory: browsing and keyword search | No external account | Fully local |
| Shared memory semantic search · full-depth graph search | An optional Google Gemini key you bring | One small embedding key; an upgrade, never a requirement |
| Voice (Ultra) | A separate provider account you bring (Grok/xAI or ChatGPT/OpenAI) | Voice's local powers — window and terminal control — cost nothing extra; only the conversational AI needs the account |
| Dictation (Pro) | Nothing — an on-device speech model | Included in every subscription; fully offline, no account, no per-word cost |
Two things never happen: the platform never resells model access with a markup, and it never meters your usage. There are no credits, no token counters, no per-run fees on top. What Anthropic (or Google, or your Voice provider) bills you is the whole model bill, on your own terms with them.
What this means for your costs
Your monthly cost has two parts:
- The subscription — fixed, per your tier (see Tiers).
- Your model usage — variable, and driven entirely by how much you dispatch. A single interactive session is cheap. A fleet of persistent agent sessions, or an aggressive routine schedule running unattended, spends more — on your Claude account, at your Claude rates.
The platform is engineered to keep the variable part down where it can be: the knowledge fabric is zero-cost by default, deterministic work runs in local services rather than burning model calls, and the orchestration doctrine explicitly steers toward the cheapest surface that does the job. But the honest statement stands: the more agent work you run, the more of your Claude capacity it uses. You control the throttle — fleet sizes, routine cadences, model choices per agent — and everything runs in the open where you can see what is spending.
What this buys you
The BYO model is not a limitation; it is the ownership posture of the whole product:
- Your keys stay on your machine. Keys and tokens live in your local configuration; they are never uploaded, and the platform's services all run locally.
- No middleman on the model relationship. Rate limits, model access, and pricing are between you and your provider. When your provider ships something better, you have it — no waiting for a platform to pass it through.
- Costs you can reason about. One fixed subscription plus a usage bill you already understand from using Claude directly.
Getting connected
Setup asks for your Claude credentials during first run, and the optional keys live in Settings where you can add them any time. The requirements page lists exactly what you need for which capability; the practical minimum is a Claude account with Claude Code access — everything else is optional until you want the capability it unlocks.
Where to go next
- Requirements — exactly which accounts and keys unlock what
- Tiers — what Pro and Ultra include
- How updates work — the other half of the subscription story
- Memory layers — where the optional Gemini key actually matters