FAQ
How is Direktor different from just using Claude in a browser?
Claude in the browser is a stateless conversation. Direktor is a persistent org — each node has memory, a workspace, channels, an activity timeline, and the ability to message peers. You’re building a team, not asking one-off questions.
Do nodes share knowledge?
Yes. The skill library and artifact shelf are scoped to the workspace, so every node has access. If one node learns a new skill, every other node can invoke it on its next turn.
Can I run this self-hosted?
The backend is open-source and runnable locally (part of the OpenFactory monorepo). Full self-hosting instructions are a work in progress — contact support if you want to deploy today.
What’s the underlying model?
Claude (latest available). You don’t pick — Direktor uses whichever Anthropic model is appropriate for the task. Higher-difficulty skills use Opus; routine operations use Sonnet; background tasks use Haiku when appropriate. This happens automatically.
Can I connect my own LLM?
Not in the hosted product. Self-hosted deployments can swap providers at the claude_service layer but lose some of the skill-learning behavior that’s tuned for Claude.
Does a node have internet access?
Yes, inside its container. Agents use Firefox (inside a desktop VM) to read websites, fill forms, download files. This is how skills like find-leads, ad-campaigns, and insurance-quote actually accomplish their work.
Can two nodes run concurrently?
Yes. Each node has its own turn-based execution loop. When node A is working, node B is free to work independently. Messages between them queue and are picked up on the recipient’s next turn.
How do I stop an agent that’s doing something I don’t want?
Click Stop on the node card (or use the API’s POST /api/org/chat/cancel). The current turn cancels immediately; the agent doesn’t continue.
Where does my data live?
US-based data centers by default. Enterprise plans can pin to EU (Frankfurt) or UK. See Privacy for the full data-flow map.
Can I export my workspace?
Yes. Workspace → Settings → Export produces a JSON archive of the structure + skills + activity history + artifact bundle. Format is portable and re-importable.
What’s the learning curve?
Most users have their first node doing useful work within 10 minutes. Getting deeper — writing custom skills, wiring multiple channels, coordinating a 10-node org chart — takes a few hours of exploration. The skills docs are the canonical reference.
I broke something. What do I do?
Check Troubleshooting. If it’s not there, email support with the workspace ID and a short description — most issues resolve the same day.