AI Agent Control Plane

AI architecture to stay agile and build operating intelligence

Context, control, and accountability for every AI agent. Leaders define where human judgement and controls remain and what agents can do. Agents reason over a growing model of your business while actions are governed, measured, and accountable to an owner, monitored, and auditable.

Foxy AI Agent Control Plane Architecture view
Context Foxy EKG Execution Knowledge Graph
Control Foxy DEAR Decision & Execution Assurance Runtime
Accountability Foxy Telemetry Monitor, Manage, ROI

Context — Foxy EKG

Build Operating Intelligence.

Each task gets the right model, decisions are grounded in the graph, and migrate to deterministic policy over time. Less inference, fewer tokens, lower cost, better performance.

@agent · accounts payableInternal SLM
Approve invoice #4471
MCPerp.invoice.lookup → ok
EKGPO matched, terms verified
✓ approved · policy99.2% · $0.0007 · 340ms
@agent · onboarding / KYCOpen Weights
Verify new account
MCPregistry · sanctions.screen
EKGentity resolved, no hit
✓ approved · inference, conf 96%$0.041 · 2.1s
@agent · claimsFrontier
Adjudicate claim #88231
MCPpolicy.coverage.check → ok
EKGprior auth on file, edge case
▲ escalated · conf 73%$0.038 · 1.9s
Inference/Policy Tracking
Claims new
30%
Onboarding
45%
Invoices mature
92%
Decision basis over time

Invoices: 41% policy six months ago → 92% today. As the EKG learns, decisions move from inference to policy.

Live decision log
14:02:11INVOICE#4471Approvedpolicy · internal SLM$0.0007
14:02:08KYCacct 90871Approvedinference 96% · open-weights$0.041
14:01:57CLAIM#88231Escalatedconf 73% · frontier$0.038

Control — Foxy DEAR

Govern every agent action.

The company sets the policy. Teams write the workflows. They compile to enforceable machine policy that DEAR checks for agent action before it runs, and creates the telemetry for repair and audit.

As written · plain languagesigned #PX-12 v2
Patient Follow-Up Workflow
  • After each visit, the agent confirms the patient's identity and checks that consent to contact is current.
  • It reviews the visit summary and care plan so the follow-up matches what the clinician ordered.
  • It drafts a follow-up message in plain language and books the next appointment in the care plan.
  • It records the interaction in the patient record and notifies the care team.
Corporate Policy — AI Agents
  • Only clinical and care-coordination staff may direct patient-facing agents.
  • Agents must mask all patient health information before sending anything to a model.
  • Agents give no clinical advice; any clinical question is held for a clinician to review and sign off.
  • Agents may contact patients only through approved secure channels, never standard SMS or email.
  • Every action an agent takes is logged and retained for audit.
As enforced · machine policyowner: Clinical Gov.
identity
  role ∈ {clinician, care_coordinator}
data
  phi.fields    mask → tokenize
  consent       require active
systems
  ehr.encounter read
  ehr.careplan  read
  ehr.note      write
tools
  appointment.schedule allow
  careteam.notify     allow
gateway / api
  portal.secure allow
  sms / email   deny (PHI)
  api.external  deny
approval
  clinical_advice require clinician_signoff
audit
  all  log → 7y (HIPAA)
Enforced live · each agent's telemetry shows the policy subset checked, pre-execution
@agent · scheduling
Book follow-up · Pt #4471
role ok · consent active · PHI masked
appointment.schedule · executed · 8ms
@agent · messaging
Text results to patient
gateway: sms · deny (PHI)
✕ blocked · routed to secure portal · 6ms
@agent · triage
Advise on symptoms
approval: clinical_advice · sign-off required
▲ held for clinician review · 7ms

Accountability — Foxy Telemetry

Customized dashboards to manage security and performance 

Real-time information for you to manage the security, performance, ad cost of your growing agent fleet. 

Agent Monitor
Technology · the controls to ensure performance & security
● 23 agents live· 20 healthy· 2 escalated· 1 broken
Agents by workflow
Accounts payable8● healthy
Patient follow-up5▲ 1 escalated
Claims adjudication6✕ 1 broken
Customer onboarding4● healthy
Model drift
low · stable 14d
Benchmark · invoice-match
model in use · 99.2%
candidate · 97.1%
Access request · 1 pending
@agent · support → crm.cases (read)
approvedenyscope ▾
AUDITDIAGNOSE
ROI Dashboard
P&L owners · cost & performance attribution to ensure ROI
67%
captured
Value capture vs FY target
$3.1M captured · $1.5M to go
return to date 4.2x
on track to target by Q4
captured remaining
By workflow · agents · process KPI · cost / mo
Accounts payable8-42% cycle$310
Patient follow-up5+18% adher.$180
Claims6-37% TAT$420
Onboarding4-29% time$140
Token use · MTD
48.2M · $1,050
Spend · weekly → projected
MTD $1,240 · proj $1,990
Foxy works within your existing IAM and alongside the apps and systems of record you already run.

Swap models, swap apps, and rationalize your systems over time. Your context, policy, and audit stay in place while the business runs.