Running & Debate

When you press Run, the board goes to work in two phases and then hands you a checkpoint. The run happens live — you can watch progress stream in as each agent starts, searches, and finishes.

Phase A — Parallel analysis

Every agent analyses your goal at the same time, each from its own adversarial angle. For each agent you get:

• an analysis of the goal,
• the claims it makes,
• the failure scenarios it foresees,
• any discoveries — new issues it surfaces, and
• questions it has for you.

If an agent decides it needs current information, it can run a live web search and re-analyse with the fresh context.

Phase B — Debate

With two or more agents, the board then debates, for up to three rounds. In each round, every agent reads the others' positions, critiques them, and revises its own stance. The debate records, per agent:

• its critiques of the others,
• its revised position,
• where it now agrees, and
• the unresolved conflicts that remain.

The debate stops early if the board converges (no new conflicts). As in Phase A, an agent can pull in live web search if it needs evidence.

Synthesis → checkpoint

Finally, Hubrix synthesises everything the board produced into a single checkpoint containing four lists:

• Checked items — assumptions and claims that were examined, each with a verdict and a confidence score.
• Failure scenarios — how things could go wrong, with likelihood, impact, and prevention.
• Discovered items — new issues, each flagged with how urgent it is and whether it needs its own agent.
• Questions for you — the decisions only you can make, with the reasoning behind each.

The checkpoint arrives with the status Needs review, and you're notified it's ready.

Credits and limits

Running a board consumes workspace credits (each agent and each debate exchange has a small cost). To keep a single session bounded, the Orchestrator enforces a per-session credit cap — if a session reaches it, the run pauses rather than spending without limit. A board also has sensible caps on the number of agents and the depth of follow-up branches.

See Billing → Credits for how credit usage works, and Reviewing results for what to do with the checkpoint.