CostPilot Policy & Rules
← Dashboard ⚖ Policy & Rules 📊 Reports 🤖 Models 🛠 Admin ⚗ Sandbox ● LIVE

Policy & Rules

Control what CostPilot may remove, protect, route, constrain, queue, skip, or block.

How CostPilot decides

Follow the controls from business context through the final model call and recorded outcome.

—Token threshold
—Complexity keywords
—Sensitive terms
—Blocking rules
6Decision stages
5Possible outcomes

Request decision order

Each stage narrows what CostPilot is permitted to do. A stop outcome prevents the model call.

1
Resolve business context and accessIdentify the project, department, human user, and agent; inactive work or unauthorized users stop here.
Identify
2
Protect sensitive informationAutomatic detectors and enabled terms may flag, force Tier 4, or block before any provider call.
Protect
3
Prune unnecessary contextWhen enabled for the payload and agent, remove redundant text and stop payloads that become effectively empty.
Optimize
4
Classify complexity and apply limitsEvaluate request length and whole-term keywords, then apply department budget throttles and agent tier limits.
Constrain
5
Select an approved modelChoose the enabled default model for the permitted tier and calculate the governed cost.
Route
6
Coordinate agents and record the decisionAgentLake applies lock, queue, or skip when agents collide; CostPilot records the final outcome for audit.
Record

Where controls are managed

Govern owns policy logic; linked pages own budgets, models, and agent settings.

Department budgets and limits → Approved models and defaults → Agent collision settings and activity →

Possible governed outcomes

Color is supporting information only; every outcome is also named and explained.

AllowProceed using the selected approved model.
RerouteUse a different tier because complexity, sensitivity, or budget requires it.
QueueWait because another agent currently owns the target work.
BlockStop before any provider call because a protection or access rule requires it.