Export metadata
Send CSV, JSON, or JSONL with provider, model, timestamp, app or API-key label, workload label, input tokens, output tokens, request count, cost, latency, status, errors, and retries where available.
Pilot audit kit
Find AI spend waste before changing production.
AIRoute runs a metadata-only review of AI usage exports and returns a savings map, workload risk labels, and next-step recommendations. The first pass does not require prompts, outputs, API keys, or customer records.
Remove sensitive fields
Do not include raw prompts, model responses, customer names, secrets, provider API keys, authentication headers, payment data, or production payloads.
Upload by invite
Use the private invite token shared by AIRoute. The current prototype accepts files up to 8 MB and stores generated reports locally for controlled early pilots.
Review the report
AIRoute returns an executive summary, top spend lanes, likely savings opportunities, and a shortlist of workloads worth benchmarking against cheaper routes.
Decide next proof
If the audit finds meaningful opportunity, the next step is observe-mode instrumentation or a narrow benchmark on approved low-risk workloads.
Keep routing separate
No production rerouting happens in the first audit. Any proxy, benchmark, or routing work should be scoped and approved separately.
Example schema
Minimum useful columns
AIRoute can normalize flexible field names, but clean labels make the report much more useful.
timestamp,provider,model,app,workload,input_tokens,output_tokens,cached_tokens,requests,cost_usd,latency_ms,status,error_count,retry_count
Ask for a metadata-only audit.
Send a contact point and AIRoute will reply with the invite token, export checklist, and upload instructions.