Resolve
An autonomous contact-center brain that handles voice and chat end to end — intent routing, authenticated tool-use against billing, CRM, and network systems, and guardrailed writes with supervisor approval for high-risk actions.
Resolve is a customer-experience product built on UAP. It does not just suggest replies for a human agent to send — it resolves interactions end to end across voice and chat. It identifies and authenticates the customer, routes intent, executes tool-use against billing, CRM, and network systems, and performs guardrailed writes, escalating only high-risk actions to a human supervisor. Every action is traced, cost-bounded, and policy-checked.
What it does
End-to-end resolution
Handles the interaction from intent to action — not draft suggestions a human still has to send.
Voice & chat
Genesys chat and Twilio voice adapters run on the same underlying brain.
Fast intent routing
A Claude Haiku router returns structured-JSON intent within a tight latency budget.
Identity & auth
A pre-processor strips PII and scores trust; an auth gate handles OTP before sensitive actions.
Guardrailed tool-use
A Sonnet resolver loops over billing, CRM, and network tools; every write is JSON-schema validated and policy-checked.
Supervisor approvals
High-risk actions — such as a credit above a ceiling — pause for a supervisor in a live approval queue.
- 1
PreProcessor
Identity, trust scoring, and PII strip.
- 2
Router
Haiku returns structured intent JSON under a tight budget.
- 3
AuthGate
OTP verification before any sensitive action.
- 4
Resolver
Sonnet tool-use loop against billing, CRM, and network.
- 5
Response
Assembles the reply; high-risk writes pause for approval.
Design targets from the reference credit scenario.
In development on UAP toward the reference scenarios: Amdocs and Salesforce connectors, the guardrailed credit flow, Genesys chat and Twilio voice adapters, a Fastify supervisor UI with a live approval queue, and a fine-tuned Router in shadow mode — targeting pilot contact-centers.
Good questions.
Is this a copilot for human agents?
No. Resolve handles interactions end to end and takes real actions, escalating only high-risk steps to a human supervisor — not the other way around.
How do you stop it making a costly mistake?
Deterministic guardrails and JSON-schema validation gate every write, with policy ceilings such as credit limits. Anything above the ceiling pauses for supervisor approval.
Can it work with our existing systems?
Yes — it connects to billing (e.g. Amdocs), CRM (e.g. Salesforce), and network systems through normalized connectors, and to Genesys and Twilio for chat and voice.