Overlay or Inlay
BUY ONE, OR BOTH
Your Tools Stay
NOTHING RIPPED OUT
Your Data Stays Yours
WE DOCUMENT HOW TO LEAVE
Authority You Set
WIDENED AS TRUST IS EARNED
Hybrid Models
OPEN + FRONTIER, CHOSEN BY US
Overlay
The Tier 1 Engineer. Investigate, Correlate, Log, Update.
Overlay does the work of a tier 1 NOC engineer. It investigates alarms, correlates them to a single root cause, gathers evidence such as packet loss and the time the fault started, checks external factors such as weather, identifies which customers are affected, logs and updates tickets, and keeps customers informed. When records conflict, it does not pick a winner: it attaches every version and escalates to tier 2 with the evidence. Like a human tier 1, it never changes the network or your systems of record.
Dozens of alarms correlated to one root cause, with the evidence attached
Affected customers identified from linked records, not searched for by hand
Existing tickets enriched instead of duplicated when your monitoring got there first
Holding message to customers straight away; full update once the facts are confirmed
FULL SPECIFICATION
AUTHORITY
Tier 1 — raises requests, never changes
ON CONFLICT
Every version attached, escalated
MODELS
Open-source, for routine work
PILOT
Available now, one incident type
Inlay
Tier 2 and Above. Analyse, Propose, Fix Within Authority.
Inlay takes escalations with a complete case file. It performs the deeper technical analysis, proposes fixes, corrects systems of record such as inventory, and makes network changes only where you have authorised it. Authority depends on the risk of the action and the impact on the customer, and you set the thresholds. Low-risk actions on non-critical customers can run automatically with a log, higher-risk ones need one-tap approval, and the highest need a named senior engineer and a rollback plan.
Deeper technical analysis on every escalation, starting from a full case file
Proposed fixes with the evidence behind them, ready for approval
Systems of record corrected at the source, only once the incident confirms the correction
Every action logged with who or what approved it, and why
FULL SPECIFICATION
AUTHORITY
Set by you, scaled by risk and impact
APPROVAL
Automatic · one-tap · senior + rollback
MODELS
Frontier models for advanced reasoning
BEST WITH
Overlay handling tier 1
Overlay + Inlay
Tier 1 and Tier 2, Working as One Team.
Buy them together and the handoff between tier 1 and tier 2 stops losing information. Overlay builds the case file and escalates; Inlay picks it up with nothing missing, confirms the correct record and corrects it at the source. The data layer detects the change, which wakes Overlay: it updates the ticket, works out the affected customers from the corrected mapping and sends the full update. The outcome is recorded in the intelligence store.
One continuous case from alarm to resolution, with no re-explaining at handoff
Conflicting records confirmed and corrected, then reflected back to the ticket automatically
Customer updates that follow the facts as they are confirmed
Every outcome recorded, so the next incident of the same kind goes faster
FULL SPECIFICATION
COVERS
Tier 1 through tier 2 and above
HANDOFF
Complete case file, every time
HUMANS
Approve what is above the threshold
LEARNING
Into the intelligence store
The Data Layer
Included With Both. Your Tools, Connected.
Overlay and Inlay both run on the data layer. It connects to your existing tools, converts their formats into one consistent shape, and links records that refer to the same thing: a customer link to its core link, an inventory ID to a monitoring ID. It keeps a synced mirror, a history of what was seen and when, and an audit trail. It takes no actions. Your own systems remain the owners of the data. Connectors, ID normalisation, linking and sync are ordinary, deterministic software; AI is used only where rules break.
Monitoring, ticketing, inventory and customer systems connected, read-only
Records across tools linked, so a fault can be traced to every customer it touches
A history of what was seen and when, plus a full audit trail
Nothing ripped out. Tools with open interfaces first, integration costs documented honestly
FULL SPECIFICATION
DOES
Connect · normalise · link · mirror
ACTIONS
None — read and sync only
OWNER OF RECORD
Your systems, always
BUILT WITH
Deterministic code, not AI
The Intelligence Store
Included With Both. What It Learns Stays Yours.
The intelligence store records what the system learns: which tool is reliable for which field, how your network's failures tend to cascade, which proposals your engineers approved or rejected, and which fixes worked. Learning comes only from reviewed outcomes, is versioned, and can be inspected and rolled back by you. It is the reason the system works best in the environment it learned in, and it is owned by you.
Learned reliability of each of your tools, field by field
How your network actually behaves when things fail
The approval history and tuned authority rules built with your team
Versioned learning you can inspect and roll back at any time
FULL SPECIFICATION
SOURCE
Reviewed outcomes only
CONTROL
Versioned · inspectable · rollback
OWNER
You
LOCK-IN
None — we document how to leave
Hybrid Models
Open Models for Routine Work. Frontier Models for Hard Cases.
We route work by difficulty. Routine correlation and normalisation for tier 1 runs on open-source models: the cheapest model that does the job well. Advanced reasoning, ambiguous cases and tier 2 escalations go to frontier closed models from labs such as Anthropic (Claude) and OpenAI (ChatGPT). You are never asked to evaluate open and closed models yourself. The model layer is built to be swapped, because prices and quality change quickly and some operators need in-country or on-premise models for sovereignty.
Open-source models for high-volume tier 1 work, keeping cost per incident low
Frontier models for advanced reasoning and escalations, where accuracy matters most
Low-confidence findings flagged for a human; consequential actions always need approval
In-country or on-premise models available where sovereignty requires it
FULL SPECIFICATION
TIER 1
Open-source models
TIER 2 + ESCALATIONS
Frontier models (e.g. Claude, ChatGPT)
CHOSEN BY
Us — managed for you
TRACKED
Cost per incident, from day one
Authority & Approvals
Included With Both. Trust, Earned Step by Step.
Operators will not hand an AI free rein over a live network, and they should not. Each product holds the authority of the human role it supports, and that authority widens only as trust is earned. Overlay holds tier 1 authority: inventory and network changes sit above tier 1, so it raises requests instead of making them. Inlay's permissions depend on two things, the risk of the action and the impact on the customer, and you set the thresholds.
Low-risk actions on non-critical customers can run automatically, with a log
Higher-risk actions need one-tap approval; the highest need a named senior engineer and a rollback plan
A critical customer on a single link gets a stricter gate than one with a redundant link
Data corrections wait for evidence and go through an approval queue
FULL SPECIFICATION
SET BY
You, per action and per customer
SCALED BY
Risk of action · customer impact
OVERLAY
Tier 1 — raises requests
INLAY
Auto · one-tap · senior + rollback
Audit Trail & a Documented Exit
Included With Both. Every Action Explained. No Lock-In.
Every action Overlay or Inlay takes is logged with who or what approved it and why, alongside the evidence it was based on. The data layer keeps a history of what was seen and when. We do not intend to lock in your data: you keep ownership of your data and your systems, and we document how to leave. The reason to stay is value that builds up over time, not a contract that traps you.
Every action logged with who or what approved it, and why
A history of what each tool reported, and when, for every incident
An audit trail you can hand to management, customers or a regulator
A documented exit: how to disconnect, and what you keep
FULL SPECIFICATION
LOGGED
Every action and approval
INCLUDES
Who · what · why · evidence
DATA OWNER
You, always
EXIT
Documented from day one
Every engagement starts
with honesty.
No sales deck. No pitch. No pre-built proposal emailed before we understand your context. A direct conversation with Bheki — and an honest assessment of fit before anything is scoped.
