Diagnostic Workbench
Join invoices, traces, application metadata, quality evaluations, and outcomes during a private engagement.
Production AI economics tools
Start with a clearer picture of the bill.
Free, private tools for measuring production AI spend, comparing alternatives, and reviewing cost controls before you change a model, provider, or infrastructure stack.
01–05 / Public tools
Choose the source data you have.
Start with assumptions, a usage export, or the production code. Each tool answers a defined question, shows its limits, and identifies the next measurement needed for a decision.
01 / Free · Browser-only
Cost-per-Outcome Calculator
Compare two production AI scenarios using retries, quality, human review, operating labor, fixed commitments, and cost per successful outcome.
Open tool 02 / Free · Browser-onlyAI Spend Snapshot
Analyze an AI usage CSV locally to see spend concentration, workload attribution, period totals, and missing measurement fields.
Open tool 03 / Free · Open sourceProduction AI Cost Review
Give Codex or Claude an evidence-based method for reviewing AI call paths, cost controls, retries, context, caching, and instrumentation.
Open tool 04 / Free · Browser-onlyInference Benchmark Planner
Compare projected cost, latency, quality, routing, retries, prompt caching, and full-response caching before running a representative local benchmark.
Open tool 05 / Free · Open sourceLLM Token Audit
Count Claude Code and Codex usage from your local session logs, then price it by model, project, session, and token type.
Open tool06–09 / Specialist tools
Public tools cover baselines. Client diagnostics handle sensitive evidence.
Quality-Cost Benchmark Lab
Turn a public scenario into a repeatable local test against representative prompts, real endpoints, and a workload-specific quality rubric.
Routing Lab
Simulate routing, fallback, quality thresholds, and cascade costs before changing a production path.
Savings Verifier
Normalize the baseline and account for changes in quality, volume, price, and workload mix.
For workload-level evidence
Turn a baseline into a production decision.
Describe one representative workload, its approximate spend, and the quality floor it must preserve.
Request a cost review →