AI control plane

Spend deliberately. Let LLM-authored position safeguards watch continuously.

Everyone may inspect the active policy. Editing models, schedules, or token budgets requires a recent Google authorization from an allowlisted administrator and creates an immutable revision for the local runtime to apply.

Active revision Not initialized —

Decision cadence — Eight paper-v3 agents per cycle

Projected decision calls — Compared with the legacy 45-minute schedule

Estimated AI spend today — Provider usage not reported

Next eight-agent cohort — Admission status unavailable

Execution safety Paper only Live routing cannot be enabled here

LLM workloads

Models and schedules

Each workload has its own model, reasoning level, schedule, timeout, and daily limits.

Whole-day cost circuit breaker

Estimated USD guardrail

$7 standard ceiling

Provider-reported input, cached-input, cache-write, output, and web-search usage is priced by the local core. A complete eight-agent cycle is admitted atomically: all eight calls run, or none do. The standard policy warns at $5 and refuses a cycle projected to cross $7.

Separate audited one-day exception

Warning: $5.00
Standard refusal: $7.00
Maximum one-day override: $10.00

Override date (America/New_York) Reason required

This action is separate from configuration revisions, expires with the selected New York date, and remains available while a formal experiment segment freezes model and cadence changes.

Always-on local automation

LLM position safeguards

No model calls

These fast jobs keep data and position rules current between deliberate LLM cycles. They are shown here for clarity and are not controlled by the model budget.

Position and exit rules: Every minute
Market/news feed refresh: Every 10 minutes
Order reconciliation: Continuous / event-driven

Protected policy

Immutable safety boundary

Execution mode: paper
Live routing: disabled
Paper-v3 agents: 8

No dashboard revision can weaken these values.

Audit trail

Configuration revisions

0 revisions

Applied	Revision	Changed by	Reason	Status
No applied configuration revisions yet.