As much Super Intelligence
as possible.

Questions

Frequently asked questions

1. 01 Can one agent think with two providers?

   Yes. An agent’s brain — the model that runs its main loop — is any catalog chat model from OpenAI, Anthropic, Google or xAI, at a thinking level you set from off to max. The loop’s supporting slots (plan, replan, reflect, strategy, extract) run on one second provider you choose, which may differ from the main loop’s. Two frontier providers, one agent.

2. 02 What happens when a provider withdraws a model?

   Model Radar probes all four providers’ model APIs every day, so the catalog never lists a dead model: whatever you can pick in chat, in a Burst squad or as an agent’s brain is a model that answers. The fleet at the foot of this page is that live catalog, by name.

3. 03 Can an agent run while I sleep — and who can stop it?

   Your workflows can: a workflow runs on a schedule (Trial 2 at ≥ 60 min · Starter 10 at ≥ 15 min · Pro 100 at ≥ 1 min · Enterprise unlimited at ≥ 1 min) or from one of 85 triggers, every run an isolated Kubernetes Job, and an agent can call a saved workflow as a single tool. An agent run starts when you hand it a job; it then works on its own — plans, calls tools in batches, reflects, asks only when it needs you — and streams into the sidebar’s live-run feed, where you can stop it mid-run from anywhere. Every run is metered and credit-gated against your balance.

4. 04 Can an agent ask me before it acts?

   Yes. Mid-run an agent may ask you a question — with choices or a short form — and pause until you answer, then carry on with your answer in hand. Before each batch of tool calls it shows a preamble of what it is about to do; after each batch it reflects and may write down a finding; it may replan. All of it streams live into the conversation.

5. 05 What does an agent carry from one run to the next?

   Lessons. Each tool keeps a Do / Don’t list of outcomes on your account — what worked, what failed and why — written by the agent in flight and by a Notekeeper after the run. A Growth Reviewer proposes new versions of the Skill and Role into a suggestion queue with evidence and a diff; you apply, edit or dismiss each one, and every version is kept — any restorable, any freezable. Training sessions let the agent practise its tools against a plan you review first. Growth off means stop changing, never forget.

6. 06 Does an agent know what day it is?

   Yes. A temporal header gives every iteration of the loop the current date and time in your time zone, so “this week”, “by Friday” and “since the last run” mean what you mean, whenever the run starts.

7. 07 Where does my data live?

   In your own private workspace on BusyBot’s Kubernetes cluster in the US. Each customer’s data is isolated to its own account (hash-tag isolation); credentials for the 295 credential types are AES-256-GCM encrypted at rest, never logged, and checked against the owning account on every use; every workflow run executes as its own isolated Job on its own node pool, and code steps run sandboxed. Conversations, files and instructions stay in that workspace as you move between models.

8. 08 How does pricing work?

   Tokens at your plan’s markup over provider list price, workflow compute by the hour beyond what your plan includes, and storage measured nightly, all settling into one credit balance. Trial is prepaid from $20 with credits good 12 months; subscriptions start at $20/month. Bring your own model keys on Starter, Pro, Enterprise.