Skip to content

Regional ISP, on-prem (name withheld)

Private AI agent platform — regulated enterprise

Deployed a multi-agent conversational system on the client's own infrastructure. Customer data never leaves their network. Agents cover dispatch, support, and field coordination.

enterprise on-prem multi-agent

The situation

A regional ISP wanted AI-powered internal tools but could not move customer data off their own network — regulatory, not hypothetical. Off-the-shelf SaaS was out. They needed an agent system that lived inside their perimeter, talked to their existing systems, and was maintainable by a small ops team.

What we did

  • Designed and deployed a multi-agent stack inside their network
  • Wired the agents into their internal ticketing and field-dispatch systems through a locked-down adapter layer that the client team controls
  • Locked management access behind their existing identity and network controls — exact mechanism documented under NDA
  • Documented the runbook so their ops team can restart, reconfigure, and extend the stack without us
  • Handoff included the entire codebase, not a black-box platform

Timeline: initial deployment in 3 weeks, followed by an ongoing monitoring retainer.

What changed

The agents handle first-pass triage on internal tickets and give dispatch a single place to coordinate across the field team. Data never leaves their network. When we push a change, it goes through their change-control process, not ours.

Relevant context

This engagement is why the WiseChef Framework product exists. The patterns we built here — isolated tenant stacks, controlled management access, LLM fallback routing, self-hosted knowledge graph — became the default architecture we now productise. The customer got a bespoke system; we got a production-validated template.

We do not publish operational details that could narrow the attack surface for a specific named client. If you need to validate the architecture for your own engagement, we cover it under NDA on the first call.