The decision engine that runs your agents.
Every minute, the Brain ingests signals, scores every possible action by ROI, decides what to do, and dispatches the right agent. You set strategy. The Brain runs execution. The agents do the work.
Without the Brain, four agents is four teams of one each fighting for the same prospect's attention. With the Brain, they're one team running one playbook — picking the highest-value work, respecting your budget, never stepping on each other. Six decision stages on a continuous loop.
- 01Sense — Ingest signals
- 02Score — Estimate ROI
- 03Decide — Resolve policy
- 04Dispatch — Route to agent
- 05Learn — Update priors
- 06Audit — Log + verify
Sense · The world, as it changes
Every change in your environment is a signal: a new lead fills a form, a deal stage advances, a renewal date approaches, a competitor mentions you on LinkedIn, a support ticket spikes. The Brain ingests them all in real time.
Sources: your CRM, your support tool, your data warehouse, every channel agent runs on, third-party intent providers. Hundreds of signal types per minute on a typical mid-market customer.
- 1.1Real-time CRM + warehouse + tool change-data-capture
- 1.2Channel events (email opens, call disposition, chat replies)
- 1.3External intent (web traffic, news, tech-stack changes)
- 1.4Internal signals (rep activity, pipeline movement, renewal risk)
Score · Every candidate, ranked
For every signal, the Brain generates candidate actions: "send Mark to outbound this account," "have Sophie confirm tomorrow's meeting," "let Alex respond to this support ticket." Each candidate scored by expected ROI: deal value × probability × urgency ÷ effort.
Thousands of candidates per cycle. Top-N kept; the long-tail discarded — so the agents only ever work on high-leverage actions.
- 2.1ROI = expected value × probability × urgency ÷ effort
- 2.2Per-agent capacity respected when ranking candidates
- 2.3Stage-aware (don't pitch close-stage deals; don't nurture late-stage)
- 2.4Continuously re-scored as new signals land
Decide · Filter by your rules
Every scored candidate is filtered through your policy matrix: per-agent autonomy levels, per-channel caps, per-account rules, kill switches, business hours, budget pacing. Anything that violates a rule never reaches an agent.
Policies are versioned, auditable, and editable by RevOps without engineering. The Brain enforces them in real time — not in a nightly sync.
- 3.1Per-agent autonomy: full auto · act+review · suggest only · off
- 3.2Per-channel + per-account overrides (VIP, do-not-contact)
- 3.3Daily / weekly / monthly action budgets enforced live
- 3.4Kill switch: stops every agent in <2s when triggered
Dispatch · The right agent, the right context
Surviving candidates dispatched to the right agent with full context: the conversation history, the related opportunity, the account's recent activity, the policy permissions for this specific action.
The agent doesn't re-fetch context every call — the Brain prepares it once, hands off cleanly, and tracks outcomes for the Learn loop.
- 4.1Context pre-assembly (history, related objects, policy)
- 4.2Agent load-balancing across capacity windows
- 4.3Cross-agent handoff (Riley → Sophie → Alex, same context)
- 4.4Failure routing — if agent declines, candidate re-queued
Learn · Outcomes feed the next cycle
Every dispatched action's outcome — replied, ignored, closed-won, ghosted — feeds back into the scoring model. Probability estimates improve. The effective ROI of each action type drifts toward empirical truth for your tenant specifically.
Your Brain learns your team's motion. What works for Sprintmore won't score the same for an insurance broker — and shouldn't.
- 5.1Per-tenant ROI model (not a global average)
- 5.2Continuous Bayesian update — no batch retraining window
- 5.3Anomaly detection (sudden drop in agent effectiveness)
- 5.4Reset + re-baseline available (e.g. new product launch)
Audit · Everything, immutably
Every decision the Brain made, every action it dispatched, every policy it enforced, every score it computed — logged immutably, exportable, queryable. You can ask "why did the Brain choose Sophie over Mark for this account at 14:32?" and get a real answer with the full inputs.
Required for compliance teams. Useful for debugging. Necessary for trust.
- 6.1Every decision logged with inputs + score + policy result
- 6.2Immutable append-only audit log
- 6.3Decision-replay tool (re-run a past decision with new policy)
- 6.4Export to your SIEM / data warehouse
Frequently asked
Set the strategy. Let the Brain run execution.
Six decision stages on a continuous loop. The Brain orchestrates; the agents do the work; you get the leverage.