# SpendLens AI > See where your AI spend goes and where it can be reduced. Track model, cache, routing and refusal costs with optional workload metadata. No proxy. No prompt storage. SpendLens AI is an AI FinOps product operated by Rangkash PTY LTD (ACN 643 391 226), based in Victoria, Australia. ## What SpendLens AI solves - Shows where OpenAI and Anthropic spend goes. - Identifies projects, models, workspaces, and API keys that deserve attention. - Finds cost-efficient models, caching opportunities, batch usage, ownership gaps, and unusual spend growth when the available provider data supports them. - Helps teams check potential savings before changing a model. ## Account Connect - Starts without an SDK or application code change. - Connects OpenAI automatically with a dedicated Admin reporting key. - Imports official Anthropic Cost and Usage CSV reports without requiring an Anthropic API key. - Keeps imported history by report date and updates matching records when the same dates are uploaded again. - Supports separate and combined views for multiple connected AI accounts, subject to plan limits. - Maps provider projects, workspaces, and API keys to applications, teams, and environments. - Shows provider cost trends, model and token mix, project drill-down, and month-on-month cost when the selected period supports them. - Surfaces model rightsizing candidates, potential prompt-cache savings, batch opportunities, service-tier signals, and ownership gaps when provider data supports them. - Encrypts stored OpenAI credentials and removes connection data when the user deletes an account. ## Runtime telemetry - Tracks OpenAI and Anthropic SDK calls without proxying AI traffic. - Captures provider, model, token usage, latency, cost, project, endpoint, and callsite metadata. - Groups calls into stable workloads and classifies their task. - Generates model-switch recommendations with confidence, quality risk, and projected savings. - Detects prompt waste signals including large context, long output, repeated templates, and high input/output ratios. - Adds deeper application-level evidence only where a team needs to validate a change. ## Compare Mode - Shortlists up to three cost-efficient, policy-approved model candidates for each classified workload. - Requires one Compare Mode option in the existing Python decorator or Node.js workload wrapper. - Runs candidate model calls inside the customer's staging application with provider keys controlled by the customer. - Keeps the original model on the application response path. Candidate failures do not replace the original response. - Sends SpendLens AI only aggregate candidate tokens, cost, latency, reliability, and error results. - Offers optional hosted quality evaluation for eligible plans. - Runs hosted quality evaluation on SpendLens AI infrastructure using SpendLens AI provider credentials, a redacted prompt template, and safe generated data. - Does not use dynamic customer values or production model responses for hosted quality evaluation. - Supports hosted quality evaluation for OpenAI, Anthropic, Gemini, DeepSeek, Mistral, and Qwen in the current version. - Uses Compare credits only for model calls and judging performed during hosted quality evaluation. Customer-side candidate calls use the customer's provider account. ## Workload economics - Measures cache efficiency, cost per useful result, and the extra cost of retries and fallbacks when the required telemetry is supplied. - Customers choose their own workload names. - Routing is performed by the customer's application or gateway. SpendLens AI measures its financial impact; it does not route requests. - Routing savings compare the chosen model with the customer's baseline model and subtract supplied classification, routing and incremental retry costs. - Missing required costs remain unknown rather than being treated as zero. Cheaper models require quality validation before switching. - A provider-specific client chooses models from that provider. Cross-provider execution requires customer-managed integrations. ## Planned: Scheduled savings - Planned for November 2026; currently an informational preview, not an active scheduling service. - Identifies opportunities to keep the same model while using eligible off-peak hours, Batch processing or Flex processing for work that can wait. - Availability depends on the model, provider, region and processing mode. Moving the start time alone does not enable Batch or Flex pricing. - Customers retain control of their scheduler and deadlines. Example savings are illustrative, not measured customer results. - See the roadmap for the planned release: https://spendlensai.dev/roadmap ## How SpendLens AI differs - Provider dashboards remain useful for native billing totals. SpendLens AI turns supported provider data into ownership, prioritization, and optimization views. - LLM observability products are useful for tracing, debugging, and evaluations. SpendLens AI starts with the financial problem and adds telemetry only when deeper evidence is needed. - AI gateways are useful for routing and request controls. SpendLens AI does not proxy or reroute AI traffic. - General cloud FinOps platforms cover infrastructure broadly. SpendLens AI focuses on model pricing, token usage, caching, batch processing, and AI model rightsizing. ## Privacy - Full user prompts are not sent by default. - template_only sampling may use system prompts or templates for workload classification. - off mode uses metadata only. - Workload classification is cached and does not run on every call. - Candidate provider keys remain in the customer's staging application and are not sent to SpendLens AI. - Dynamic prompt values and production responses are not used for hosted quality evaluation. ## Plans - Free: 1 provider connection, 30-day provider history, and 10,000 runtime events per billing cycle. - Starter: 2 provider connections, 90-day provider history, and 100,000 runtime events per billing cycle. - Pro: 5 provider connections, 12-month provider history, and 1,000,000 runtime events per billing cycle. - Growth: 15 provider connections, 24-month provider history, and 5,000,000 runtime events per billing cycle. - Pro includes $40 in Compare credits each month. - Growth includes $100 in Compare credits each month. ## Canonical pages - Home: https://spendlensai.dev/ - Problem and solution: https://spendlensai.dev/problem - How it works: https://spendlensai.dev/how-it-works - Account Connect: https://spendlensai.dev/account-connect - Alternatives and comparisons: https://spendlensai.dev/alternatives - Savings calculator: https://spendlensai.dev/calculator - Pricing: https://spendlensai.dev/pricing - FAQ: https://spendlensai.dev/faq - Blog: https://spendlensai.dev/blog - Roadmap: https://spendlensai.dev/roadmap - Documentation: https://spendlensai.dev/docs - Quickstart: https://spendlensai.dev/docs/quickstart - Python SDK: https://spendlensai.dev/docs/python-sdk - Node.js SDK: https://spendlensai.dev/docs/node-sdk - Workload economics: https://spendlensai.dev/docs/workload-economics - Compare Mode: https://spendlensai.dev/docs/evaluation-campaigns - OpenTelemetry: https://spendlensai.dev/docs/opentelemetry - API keys: https://spendlensai.dev/docs/api-keys - Privacy: https://spendlensai.dev/docs/privacy - Sample report: https://spendlensai.dev/sample-report - Contact: https://spendlensai.dev/contact - Human-readable product facts: https://spendlensai.dev/llms - Plain-text product facts: https://spendlensai.dev/llms.txt ## Important limitations - Savings are estimates, not guarantees. - Model changes should be evaluated for quality, latency, safety, privacy, residency, and contractual requirements. - Provider reporting APIs and CSV exports do not expose every metric for every account. - Account Connect identifies areas to investigate. Runtime telemetry provides deeper evidence for application-level changes. - The current version does not automatically rewrite prompts. - SpendLens AI does not collect payment card details; Stripe Checkout and Customer Portal handle billing.