Key facts
| Router model | model="plugsky-fusion" applies chain and rules per request (live) |
| Chain | Cost-first default escalates micro to pro to frontier on hard prompts |
| Throughput | Fair-use request limits scale by tier, up to 10K+ RPM on Enterprise |
| Rules | Custom rules can key on tokens, tools, language or workspace context |
| Residency | Pin workspaces or requests to GCC, EU, US or APAC; on-prem and air-gapped available |
| Audit | Model, strategy, rule and request metadata logged and exportable to SIEM |
| Cost shape | Flat plans with fair-use RPM; no per-token charges on self-serve |
| Roadmap | Classifier routing /v1/plugsky/route (model=auto) is coming soon |
TL;DR
- Run the bulk of care traffic through a cost-first chain.
- Escalate complex cases automatically instead of staffing for peaks.
- Use rules for language, queue and business-hours behaviour.
- Pin subscriber-data workloads to the required jurisdiction.
- Keep every routing decision auditable at telco scale.
How it works, step by step
- Split traffic into buckets: self-service enquiries, care chat, assurance triage, complex retention cases.
- Assign a strategy per bucket and a default chain for anything uncategorised.
- Add rules where markers are reliable — prompt length, tool use, language or originating queue.
- Load-test the peak hour and confirm the fair-use request rate for your tier covers it.
- Pin each workspace to the required region and wire routing logs into your observability stack.
- Review escalation rates monthly and tune thresholds with quality data.
Try it yourself
Volume economics for care traffic
Telecom care volume dwarfs almost every other industry: balance checks, plan questions, outage queries, billing disputes. Sending all of it to a frontier model is economically indefensible, and sending all of it to a weak model is a customer-experience risk. Fusion splits the difference — a cost-first chain answers the routine majority and escalates the complex minority automatically.
Because self-serve plans carry no per-token charges, the optimisation target is quality and latency rather than cost accounting. Tune the chain until escalation happens for the right reasons, then leave it alone and monitor.
Rules that reflect how telcos operate
Telco traffic has structure you can exploit. Business-hours traffic may deserve a stronger default; out-of-hours self-service can be cheaper. Language heuristics can route to models that perform best for that market. Tool-heavy requests — plan changes, order status — can be pinned to a capable tier because a wrong answer costs more than a strong model call.
- Overrides: fixed model names bypass routing for regulatory or determinism-bound flows.
- Escape hatch: pin a premium model for retention queues where quality directly protects revenue.
- Roadmap: classifier-based selection arrives with POST /v1/plugsky/route (model=auto), coming soon.
Sovereignty, latency and oversight
Subscriber data invites regulation. Pin workspaces to the region your authority requires — GCC, EU, US or APAC — or deploy in your VPC, on-prem or air-gapped where data must not leave the estate. Data never leaves the pinned region.
At scale, oversight cannot be manual. Every request logs the model, strategy, rule, tokens, latency, status, key ID, project and region, and those events stream to Splunk, Sentinel, QRadar or Chronicle. Load-test with the latency tester, capacity-plan against fair-use request rates up to 10K+ RPM on Enterprise, and treat routing changes as change-managed configuration.
Honest comparison
| Concern | Plugsky Model Fusion | Single frontier model | Vendor telco suite |
|---|---|---|---|
| Unit economics | Cost-first chain with escalation | Premium on every call | Contract-dependent |
| Throughput | Fair-use tiers to 10K+ RPM Enterprise | Provider limits | Suite limits |
| Rule control | Custom rules on tokens, tools, language | None | Vendor-defined |
| Residency | Regional pinning, on-prem, air-gapped | Provider regions | Vendor regions |
| Audit | Model, strategy and rule per request | Basic logs | Suite logs |
Frequently asked questions
Can fusion handle peak-hour telco volume?
Enterprise tiers carry fair-use request limits up to 10K+ RPM with the same flat rate on every model. Load-test against your peak and use parallel tool calls to cut round trips.
How do we keep subscriber data in-country?
Pin each workspace or request to the required region — GCC, EU, US or APAC — or deploy on-prem and air-gapped if data cannot leave your estate.
Does routing help with multilingual care?
Rules can key on language or originating queue so each market gets a model that performs well for it. Validate with your own evaluation set.
How is quality controlled at volume?
The chain escalates hard prompts automatically, and the decision log shows escalation rates per queue, model and rule so you can tune thresholds with data.
Can we force a model for certain flows?
Yes. An explicit model name bypasses routing, which is how we recommend handling regulated or determinism-critical flows.
What does an audit trail contain?
Every request logs model, strategy, rule, tokens, latency, status, key ID, project and region; events can be exported to your SIEM.
How do we pilot this?
Pick one queue, run it on a cost-first chain for a month, and compare escalation rate, resolution quality and latency before scaling across care and assurance.