◆ Autonomous Specialist
Watches utilization, forecasts demand, and applies scaling and instance-mix changes within bounds. It backtests each change against historical load and acts on its own forecasts, holding headroom for peak while trimming idle capacity.
Memory
Working Current utilization snapshot and the change under evaluation.
Episodic Past scaling events and their cost/performance outcomes.
Semantic Service SLOs, instance pricing, reservation commitments.
Store Time-series store + file-based memory tool
Orchestration
MCP
Harness · Managed Agents: scheduled + event-driven; sandboxed forecasting backtests.
Tools
{ } Cloud billing + usage API API { } Autoscaler config API ›_ Demand-forecast sandbox Code exec
Evals & guardrails
- Changes above a spend/risk threshold require an oversight-agent gate before commit.
- Backtest every right-sizing change against historical load before applying.
- SLO-protection guardrail: never scale below the headroom needed for peak.
Frontier edge
- ▲World-model simulation: forecasts demand and simulates instance-mix and reservation changes against modelled load before applying them.
- ▲Causal cost attribution: counterfactual 'which service actually drove the bill spike', not just which line item correlated with it.
- ▲Proactive right-sizing: pre-positions capacity ahead of a predicted month-end or market-event peak rather than reacting to throttling.
In numbers
−19%
Infra spend reduced
0
SLO breaches from under-provisioning
Handoffs
Fed by ← Incident Response Agent