Cortex
◆ Cortex / Foundry / Smart Routing & Enrichment

Other routers route prompts. Cortex routes the work.

Route every request by Work Unit, not just tokens. Right model, proven context, cost, latency, and risk limits on every call.

  • Frontier models only when the work earns it
  • Every prompt carries your org's context
  • Cost, latency, and risk limits enforced on every call
Talk to the team
Pairs with Context & Memory Governance & Security
Cortex Router
Best-value route

Route the work. Not the model menu.

Cortex routes every request by Work Unit, not raw prompts. Company and team policy, cost, latency, risk, and your quality bar pick the path that clears the bar for the least cost per outcome.

Foundry · Router
Work Unit Triage this support escalation, with prior resolutions Enriched
Task · support triage Actor · L2 support Gate · support-gold ≥ 90% Prior · cortex-triage-v3 · 96% accepted Policy · cost · latency · risk
GPT-4o Frontier
Cost$0.0416.8× Latency1,180ms3.1× Evalpass

Clears the bar, but frontier price for a routine triage path. Context already in the enrichment layer.

Enriched with prior tickets runbook PII policy
cortex-triage-v3 Best value
Cost$0.006base Latency380msunder cap Eval94% · gate 90%

Owned model that already passes your support gold set. Cheapest path that clears the quality bar inside policy.

Enriched with prior resolutions gold-set peers team memory risk: PII-safe
Claude Sonnet Fallback
Cost$0.0183.0× Latency690ms1.8× Evalpass

Solid mid-tier path if the owned model is cold or the Work Unit drifts outside its proven scope.

Enriched with prior tickets escalation graph latency budget
  • Best value, not biggest modelThe cheapest path that passes your evals, never the frontier model by default.
  • Enriched before it runsEvery candidate sees the same org context. The route picks cost and fit, not a bare prompt.
  • Inside your policyEvery route respects your cost, latency, and risk limits, and every permission.
Prompt enrichment

Sparse prompt in. Your whole org behind it.

At run time, Cortex searches your company context graph and memory stores that Atlas builds. The task, the artifacts, and the history attach before the model runs. Ten words in, a grounded answer out, at a fraction of the cost.

Cortex Enrichment
Beyond the prompt

Other routers route prompts. Cortex routes the work.

Cortex picks the model using signals no prompt-based router has: the task being done, the person's role and permissions, and which model produced accepted work on that task before.

Foundry · Routing signals
Prompt-based router Triage this support escalation
Prompt textToken length
Cortex Triage this support escalation
Task: support triageActor: L2 support · permissionsPrior outcome: cortex-triage-v3 · 96% acceptedBusiness object: Acme accountMemory: 3 similar resolutionsPolicy: cost cap · PII scope
Route: GPT-4o · $0.041
Route: cortex-triage-v3 · $0.006 · passes eval Best value
How we measure

Cortex Router benchmarks

Public measurements.

Cortex Router vs Sol Sol quality with 27% savings. Cortex Router vs Not Diamond 91% of Opus quality with 40% savings.

See Smart Routing & Enrichment
on your own data.

We'll scope it as part of Cortex Foundry on your stack in days.

Talk to the team Explore Foundry