Cross-Domain Orchestrator

Custom agent·policy router + tools

Connected

T1 · Inline

Full fidelity·The agent's model and tool calls route through Sentry's gateway.

Span-level analysis of every conversation this agent runs – the token economics behind each answer and the tool-call orchestration underneath it. Levers span model routing, semantic model routing, prompt caching, context pruning, schema projection, request batching, retrieval (RAG) tuning, speculative decoding and token budgeting – cutting cost and latency without moving quality.

Token economics

Cache hit rate

41%

cached ÷ input tokens

Context utilisation

64%

avg window used

Tool-result share

17%

of total tokens

Cost / resolved task

$0.09

blended

Weekly token composition920k total

Input47%Cached20%Tool result17%Output16%

Tool-call orchestration

17,510 calls/wk · 16% redundant · 1 cacheable

Tool / span	Calls/wk	Avg tokens	p95	Redundant	Lever
EEricsson NMS API	5,210	1,700	1.6s	12%	Parallelizable
NNetcracker API	4,800	2,200	2.1s	24%	Over-fetch
NNokia NSP API	3,600	1,400	1.3s	6%	Healthy
SSAP connector	2,100	2,600	3.0s	31%	Cacheable
SServiceNow API	1,800	1,200	1.1s	5%	Healthy

Optimization levers

Recommended

Data orchestrationParallel fan-out

Parallelise domain lookups

Core, OSS/BSS and Transport queries run sequentially today. Fan them out concurrently to cut p95 latency and timeout retries.

−150k/wk · −1.4s p95 · quality +1%

Data orchestrationResult caching

Cache the legacy ERP connector

ERP records are read-heavy and stable. A 15-minute TTL cache removes 31% redundant calls and their tool-result tokens.

−90k/wk · −1.8s p95 · quality +0%

Token efficiencySchema projection

Trim OSS/BSS payloads

OSS responses include unused fields. Project to only the fields the agent reasons over to cut input + tool-result tokens.

−130k/wk · −0.4s p95 · quality +0%

Applied

Pass domain deltas only; keep full inventory in a cached tool−65%

after

4,900 / 14,200