CadreGPT
Plans & Pricing
Sign In Start Free Trial
Algorithmic Budget Protection & Isolation

Two-Phase Financial Invariance

Eliminate runaway LLM bills forever. CadreGPT enforces mathematical two-phase credit reservations, SERIALIZABLE database isolation, and hard spending caps across every autonomous task turn.

Phase 1: Pre-Turn Reservation

Before prompt tokens or tool operations dispatch, the engine computes a worst-case credit allotment and places a temporary hold on your workspace balance. If the balance is insufficient, execution halts safely before incurring external debt.

Execution: Sandboxed Metering

Agents query foundation models and execute authorized tools within an isolated runtime. Exact prompt tokens, completion tokens, and tool runtimes are captured in real-time telemetry streams.

Phase 2: Delta Settlement & Release

Upon turn conclusion, the actual token consumption is calculated using provider rates. Only the exact consumed credits are deducted; the unused reserved delta is immediately unlocked back into your active balance.

SERIALIZABLE Ledger Isolation

PostgreSQL multi-tenant ledger rows utilize SELECT FOR UPDATE locking under SERIALIZABLE transactions to prevent race conditions and double-spending across parallel worker swarms.

Hard Monthly Spending Caps

Define immutable financial ceilings. Any automated recharge or scheduled run that would breach the configured ceiling is cleanly suppressed with proactive administrative notifications.

Pinned Unit Currency ($0.001)

1 Agent Credit = $0.001 USD. Clear, predictable unit economics make budgeting and forecasting straightforward across engineering, marketing, and operations teams.

Knowledge Base & FAQs

Frequently Asked Questions: Financial Invariance

Learn how our two-phase metering architecture guarantees budget safety and prevents runaway costs.

01.What is two-phase financial invariance in AI agent operations?

Financial invariance is an algorithmic guarantee that an organization’s compute balance cannot drift into uncollected deficits or trigger uncapped provider invoices. Under CadreGPT’s two-phase model, a maximum potential credit ceiling is reserved prior to dispatch. When execution completes, the exact token cost is computed, the exact consumed amount is deducted, and the unused reservation delta is atomically unlocked.

02.How does CadreGPT prevent unexpected runaway token expenses?

Traditional agent loops make direct calls to provider API keys without budget governors, leading to multi-thousand-dollar accidents if an agent gets trapped in an infinite loop. CadreGPT enforces hard token cost ceilings per execution turn, hard monthly spending caps per workspace, and automatic loop detection.

03.What happens if a worker node crashes mid-execution during a reservation?

Our BullMQ distributed queue and Redis Redlock architecture maintains heartbeat leases. If a worker fails, the heartbeat expires and the two-phase credit engine automatically releases any reserved credits back to the workspace ledger, guaranteeing zero financial leakage.

04.How are Agent Credits calculated across different LLM providers?

Agent Credits are pegged at 1 credit = $0.001 USD. When agents invoke Google Gemini, Anthropic Claude, or OpenAI models, prompt and completion tokens are metered against live provider unit prices, mapped directly to credits with zero hidden markups on direct credit billing.

05.Can I configure hard spending caps without disabling live agents?

Yes. Workspaces can establish hard monthly limits and auto-recharge caps. If a limit is approached, automated warning alerts trigger. If the limit is reached, auto-recharges are cleanly suppressed and newly queued runs pause safely in an awaiting approval state rather than failing catastrophically.

Protect Your AI Compute Budget

Deploy autonomous workflows with guaranteed financial invariance and zero unexpected overage invoices.