Products
Pricing Blog Join the waitlist
← Back to BehaviorGraph
Flagship use case

Org context for agent routing.

AI agents doing enterprise work hit the same wall: they do not know who actually does what. The org chart and the documents both miss how work really happens. The Large Behavior Model supplies the missing layer, queryable at runtime, so your agent assigns, approves, and escalates like someone who knows the building.

Three dimensions, one call

Every candidate is scored on who knows the work, who is safe to route to, and who can actually take it on right now.

🎯

Relevance

Who truly knows this domain, read from who peers actually rely on and who owns the living documents. Not who has the loudest title.

🛡️

Trust

Who is safe to route to: peer endorsements, past delivery, and approval rights, so routine work never lands on an executive's desk or an unproven one.

⏱️

Capacity

Who is not overloaded right now, from live load and response patterns, with a qualified backup returned when the primary is at capacity.

One call, a defensible decision

Send a task, question, or exception. BehaviorGraph returns a ranked, capacity-aware answer with the reasoning attached, so the decision can be audited later.

  • Primary owner: the person who actually handles this type of work
  • Qualified backup when the primary is at capacity or away
  • Who to avoid, and why: overload, conflict of duties, or wrong domain
  • The real escalation path, filtered by approval rights, with an SLA
  • A reason for every recommendation, logged for audit
// Agent detects an invoice stuck past SLA
POST /v1/agents/route

{
  "query": "invoice stuck past SLA, APAC vendor",
  "domain": "finance/accounts-payable",
  "action": "exception"
}

───────────────────────────────────

// BehaviorGraph returns
{
  "primary": "FIN-04",
  "reason": "owns this contract type, available",
  "backup": "FIN-09",
  "avoid": "FIN-01 (at capacity, 92h/wk)",
  "escalation": "24h → Dir. Finance"
}

Escalate with context. Govern by design.

When a decision needs a human, the right person gets it with the reasoning, alternatives considered, and permission boundaries already attached. And every recommendation is logged.

  • Routes within your approval structure, never around it
  • Knows when to stop and ask a human, and which human
  • Every decision traceable: who was chosen, why, what was skipped
  • Metadata only, anonymized person codes, privacy by design
// Decision log, written automatically
{
  "decision": "routed to FIN-04",
  "considered": ["FIN-01", "FIN-04", "FIN-09"],
  "skipped": "FIN-01: at capacity",
  "policy_check": "within approval structure",
  "human_in_loop": "not required at this tier",
  "audit_id": "rt_88412"
}

Tested on real tickets, not a demo dataset

Results from current research runs, validated externally. The full validation surface, including what is partial, is in the model overview.

🧠

Recent expertise beats seniority

On real tickets, who handled this type of work lately predicts the resolver better than rank or tenure. Routing by org chart sends work to the wrong desk.

🔍

A reason on every decision

Each recommendation returns its evidence: peer reliance, document ownership, load. Auditors and owners can see exactly why the agent chose who it chose.

Two lines to add routing context to your agent

REST API or MCP (Model Context Protocol). Works with LangChain, CrewAI, AutoGen, or your own orchestration.

POST /v1/agents/route

Before your agent takes a consequential action, ask BehaviorGraph who should handle it, who's available, and what the escalation path is.

Returns: primary, backup, avoid list, escalation path, reasoning
MCP behaviorgraph/route

If your agent framework supports MCP tools, BehaviorGraph registers as a tool your agent can call natively. No custom integration code needed.

Compatible with: Claude MCP, LangChain MCP, any MCP-compatible runtime
GET /v1/context/person/{id}

Look up a person's behavioral profile: what they're relied on for, their current load, team centrality, and escalation path.

Returns: domains, relevance scores, load, team connections, escalation chain
GET /v1/context/team/{id}/health

Before routing work to a team, check their health: bottlenecks, overloaded members, available capacity, and bridge connectors.

Returns: team load, bottleneck nodes, available capacity, bridge connectors

Give your agents the org, not a routing table

The Large Behavior Model keeps a live, queryable read of how your organization actually works: ownership, trust, capacity, and escalation paths that update as the org changes.

Get a demo Read the model overview