Agent/MCP Audit Sprint

LLM spend estimate to paid scope packet

OpenRouter Cost Calculator

Estimate one OpenRouter or LLM API workload from manual counts or sanitized usage rows, see whether it fits the USD $99 OpenRouter Cost Sanity Check, USD $299 OpenRouter Agent Cost-Control Review, or USD $1,000 AI Cost Spike Emergency Sprint, then copy a sanitized intake packet. Pricing presets were checked against the OpenRouter models API on 2026-06-30 and remain editable snapshots, not a billing guarantee; verify provider pricing before accepting scope.

Run calculator Actual cost check Balance guardrail 402 brownout LiteLLM budget TokenMeter quick audit Agent cost review Emergency sprint
Free toolEditable OpenRouter cost estimate
QuickUSD $99 OpenRouter Cost Sanity Check
FocusedUSD $299 OpenRouter Agent Cost-Control Review
EmergencyUSD $1,000 AI Cost Spike Emergency Sprint
Start RulePayment only after written scope acceptance

Calculator

Turn model usage into a review packet

Use sanitized counts only: request volume, model mix, token estimates, cache-read share, retry rate, and tool-call fanout. You can paste redacted JSON, JSONL, or CSV usage rows to fill the calculator. Do not paste private prompts, customer data, API keys, provider account IDs, raw traces, or billing screenshots with sensitive values.

Monthly$0.00 estimated monthly model cost
Daily$0.00 estimated daily model cost
FitUSD $99 OpenRouter Cost Sanity Check
GuardrailPayment only after written scope acceptance; calculator output is not a payment request.

Usage import

Paste sanitized OpenRouter JSON, JSONL, or CSV rows

Supported fields include model, input_tokens, prompt_tokens, output_tokens, completion_tokens, cache_read_tokens, cached_tokens, status, error, retry, and tool_calls. Rows are scaled to a monthly estimate using the sample window below.

Usage import has not been applied.

Open GitHub intake

Use balance as a signal, not permission

A credits or key-limit snapshot should inform the UI, but the dispatch gate still needs a reservation before the next model call and an actual-cost reconciliation after the generation finishes.

Open balance guardrail Open actual-cost reconciliation Open key docs
Account credits: track purchased and used credits separately from local monthly budgets.
Key limits: treat remaining limit, reset policy, and usage as the current key scope, not a global account truth.
Generation cost: reconcile token counts and usage after completion, including streaming and provider fallback cases.
Stale state: block or degrade expensive calls when credits, key limits, or generation metadata cannot be trusted.

Routing

Pick the package from the cost shape

The calculator routes small estimates to a quick API cost scan, recurring workflow waste to the focused review, and already-spiking bills to the emergency sprint. The final package is still agreed in writing before any payment request.

USD $99 OpenRouter Cost Sanity Check: a small model mix, launch-pricing, or provider-cost sanity check.
USD $299 OpenRouter Agent Cost-Control Review: one OpenRouter workflow with context bloat, model-routing drift, cache misses, retry loops, actual-cost reconciliation, or tool-call fanout.
USD $1,000 AI Cost Spike Emergency Sprint: a live bill spike, fast burn rate, failed launch pricing, or urgent containment need.
Evidence boundary: use model ids, token totals, counts, retry rates, and redacted summaries, not secrets or raw private traces.

Cost drivers

What the paid review checks after intake

A calculator estimate is only the first pass. The paid review looks for the product and engineering levers that change recurring cost: routing, cacheability, context selection, retries, and feature-level attribution. Web-search charges, cache-write fees, provider markups, and custom discounts are intentionally left as manual adjustments.

View $299 focused review View $1,000 emergency sprint
Whether cache-read savings are actually reachable with stable prompt prefixes and reusable context.
Whether expensive models are reserved for the work that needs them, with cheaper default paths for routine steps.
Whether retries, repair loops, tool fanout, and browser/coding-agent loops have hard spend caps.
Whether product pricing, free-tier assumptions, or customer-level budgets survive realistic monthly usage.
Whether logs can attribute cost by feature, customer, model, tool, and run without exposing sensitive data.