Pages
Three planes, two clocks, and usage to bill
Seconds for signals, days for meaning, and a governed hop in between.
Exosphere Agentic runs on Azure and integrates with the Microsoft AI platform, deployed into the customer's own tenant — so context, data and governance stay where they belong. Network state changes in seconds, while the meaning of that state changes on change-order timescales. Exosphere Agentic keeps each on the clock it suits, and attaches governance to the hop where an event becomes an action.
Updated
Two clocks, one decision
BSS and OSS data move at different speeds, and a good design respects both.
- BSS is transactional. Customers, offers, orders and bills change on human timescales, and the record of what happened is the asset.
- OSS is current state. Cells, ports, slices and network functions run deep, topology keeps changing, and signals arrive as periodic counters, streaming telemetry and asynchronous alarms.
Dispatch needs current state. The credit needs the record. Those are two planes, and Exosphere Agentic treats them that way.
TM Forum draws the same line. IG1219 separates fast closed loops, which act on real-time inputs, from slow closed loops, which act on history and feedback. The GB922 Closed Loop pattern lets loops run fast or slow, and nest or work as peers.
Three planes at the BSS/OSS crossing
flowchart LR
subgraph NET["NETWORK AND OSS SOURCES"]
N1["Alarms<br/>TMF642"]
N2["Counters and KPIs<br/>3GPP PM · TMF628"]
N3["Thresholds<br/>TMF649 · TS 28.550"]
end
subgraph FAST["FAST PLANE · SECONDS"]
F1["Event streams<br/>windowing and thresholds"]
end
subgraph SLOW["SLOW PLANE · DAYS"]
S1["Telecom ontology<br/>Resource · RFS · CFS · Product · Customer"]
S2["Commercial terms<br/>Agreement TMF651 · SLO TMF657"]
end
subgraph ACT["ACT PLANE · GOVERNED"]
A1["Assurance agent"]
A2(["Human approval gate"])
A3["Governed TMF calls<br/>TMF656 · TMF621 · TMF724 · TMF681"]
end
BIL["Billing and care<br/>credit visible via TMF678"]
AUD["Audit record"]
N1 --> F1
N2 --> F1
N3 --> F1
F1 -->|"condition met"| A1
A1 -->|"who is affected"| S1
S1 --- S2
S2 -->|"what we owe"| A1
A1 --> A2
A2 --> A3
A3 --> BIL
A3 --> AUD
classDef exos stroke:#0077B6,stroke-width:2px
classDef human stroke:#009688,stroke-width:3px
classDef std stroke:#005F8A,stroke-width:1px
class S1,A3,AUD exos
class A2 human
class N1,N2,N3 stdEach plane does only what it is good at
| Plane | Clock | What it holds | Its job | In a Microsoft deployment |
|---|---|---|---|---|
| Fast | Seconds | Alarms, counters, KPIs, thresholds | Windowing, thresholding, detecting a condition | Event streams in Microsoft Fabric Real-Time Intelligence |
| Slow | Days | Topology, service decomposition, resource-to-service and customer-to-product bindings, SLA terms | Meaning: who depends on what, and what was promised | The telecom ontology, bound in Fabric IQ |
| Act | Each decision | Proposals, policy, approvals, governed calls | Execute, route or block, then call TM Forum Open APIs and write the audit record | Exosphere, the judgment layer inside Exosphere Agentic, through Azure API Management |
Semantics stay out of the real-time path, so the ontology never has to move in seconds, and nothing in the fast plane waits on a refresh. Microsoft Fabric supplies analytics and context. The execution path runs through Exosphere.
Intent sets the goal, and the hop to action is governed
- Intent stays out of hard real time. TM Forum positions intent handlers as setters and adjusters of goals for faster loops, not as real-time controllers (IG1358). Exosphere follows the same split: policy and intent live above the fast loops.
- Authorisation attaches at the hop. A stream condition raises a signal. Only Exosphere holds write scopes, so the step from signal to action is exactly where identity, authority and audit attach.
- Least privilege for loop data. TM Forum's NaaS closed-loop data map (IG1224, aligned to ETSI ZSM) assigns standard APIs to each loop input: TMF642, TMF649, TMF628, TMF656 and TMF657 for events, TMF639 for inventory, TMF640, TMF641 and TMF664 for configuration. It asks for that data to be exposed with authorisation and least privilege.
- The 3GPP side fits the same shape. TS 28.533 builds management on producer and consumer management services, with exposure governance for consumers outside the operator. Performance data arrives as a stream (TS 28.532 and TS 28.550) or as files, with threshold-crossing notifications.
Usage to the bill
flowchart LR
CHF["3GPP CHF and CDRs<br/>TS 32.240 · TS 32.298"]
RES["Resource"] -->|"TMF771 Resource Usage"| RU["Resource usage"]
SVC["Service · CFS or RFS"] -->|"TMF727 Service Usage"| SU["Service usage"]
PRD["Product"] -->|"TMF635 Product Usage"| PU["Product usage"]
SPEC["TMF767 Product Usage Catalog"] -.->|"specifies"| PU
CHF -.->|"usage records via mediation"| PU
RU -.->|"can influence"| SU
RU -.->|"can influence"| PU
PU -->|"rated in charging"| ABR["Applied billing rate"]
ABR -->|"TMF678 Customer Bill"| BILL["Customer bill"]
BILL -->|"TMF666 Account"| ACC["Billing account"]
classDef std stroke:#005F8A,stroke-width:1px
classDef sor stroke:#7A8DA5,stroke-width:1px
class RU,SU,PU,SPEC std
class ABR,BILL,ACC sorUsage, split the way SID splits it
- Three usage APIs, three SID entities. TMF635 is now Product Usage. Service usage sits with TMF727, the newest of the three, and resource usage with TMF771. Usage specifications live in TMF767. This mirrors SID's Product, Service and Resource Usage entities.
- Usage arrives slice-aware. In 3GPP converged charging (TS 32.240) the charging function, account balance management, rating and the charging gateway work as one system. For 5G data (TS 32.255) the session management function triggers charging, and the charging data for each session names the network slice instance it belongs to.
- From rating to the bill. Rated product usage becomes an applied billing rate on the customer bill (TMF678), held against the billing account (TMF666).
Why it matters. The same entity chain that answers who is affected also answers what did they use, and what did we charge. A credit, a re-rate and a usage query all resolve through one model.
Energy per service is a natural extension on the same decomposition. Since Release 19, 3GPP CDR parameters (TS 32.298) can carry network-slice energy information, and Exos Systems models the attribution from resource to service to customer as part of the engagement path.
Take the next step
See the loop run on synthetic accounts, or walk through the ontology with Exos Systems.