Zetalens connects to your billing APIs and surfaces cost data by provider, model, team, and seat. The initial connection takes 15 minutes. The data available to your CTO, Engineering leads, and Finance team is immediate and does not require ongoing maintenance.
Why Zetalens
Direct API providers and wrapped SaaS tools in one view with consistent spend data, trend lines, and attribution. No instrumentation. No code to deploy.
Map billing data to your team structure so squad leads, service owners, and platform teams see exactly what they are consuming. No tagging conventions to enforce.
Department-level cost allocation, 12 months of unified spend history, and exportable board-ready summaries. Forecast AI spend with the same discipline as every other budget category.
Features
See your spend. Attribute it to the team and seat that consumed it. Stay in control of where it goes next.
One dashboard, every provider. Board-ready summaries on demand. Per-employee normalization that turns raw spend into a defensible benchmark.
Every provider in one view with consistent data. Provider-level spend, month-over-month trend, API costs versus SaaS subscription split.
Michael sees 90 days of spend across six providers on one screen. Anthropic up 34% on a new feature. Two-page summary to the CFO before the prep call.
Cost broken down by model within each provider. GPT-4o versus mini, Opus versus Sonnet. The data you need to choose models on cost and capability.
70% of Anthropic spend on Opus for a workflow scoped as a prototype. Switch to Sonnet, $8,400/month saved, no measurable quality loss.
Total AI spend by provider and business function, month-over-month trend, model-level breakdown. Exports as clean PDFs for executive presentation.
Monthly export to the CFO scheduled for the first business day. Total spend, top three drivers, 12-month trend. Half-day of work, now zero.
Per-employee and per-department normalization. Tells you whether spend is efficient, not just what it totaled.
Legal at $340/employee/month, 4x company average. Two heavy Harvey users plus 14 dormant Jasper seats. Removed unused, justified the rest.
Attribution at every level your organization actually operates: squad, API key, SaaS seat. Then export the allocations to the systems Finance already uses.
API spend mapped to engineering teams using existing key structure. Squad leads see their own costs without custom tagging infrastructure.
Search squad lead notices Anthropic spend up 60%, traces to a new reranking model, fixes it before billing closes. Sarah hears about it from the squad, not the invoice.
Cost attributed to the team or service that made each call. Where keys are shared, Zetalens recommends cleaner key structure.
Three squads share a single OpenAI key. Zetalens flags the gap, platform team adds squad-scoped keys, clean attribution within 30 days.
Seat-level activity for Cursor, Copilot, Claude.ai Teams. Last active date and utilization tier for every seat. Filter for inactive seats before renewal.
Of 48 Cursor seats, 11 inactive in 45 days. Removed at renewal, $4,840/yr saved, remaining 37 justified with usage data.
CSV exports for FP&A and ERP, PDF exports for board decks and audit. Templates structured to match standard cost allocation formats.
Controller exports allocation to CSV, drops it into the FP&A model. PDF summaries to department heads automatically. 90 seconds, replacing 4-6 hours.
Find out when spend moves, not when billing closes. Forecast month-end on real consumption. Walk into renewals with vendor data Finance can defend.
Configurable thresholds by provider, service, team, or portfolio. Daily monitoring with alerts via email or Slack when a limit is crossed.
Tuesday alert: OpenAI up 210% from baseline. Batch job deployed to production by mistake. Stopped within two hours, not at month end.
Month-end projection by provider based on daily run rate, adjusted for day-of-week patterns. Updates in real time throughout the month.
Day 10: tracking $52K against $45K budget. Anthropic the variance driver. Engineering finds the feature, adds a rate limit, month closes at $47.8K.
Vendor plan data alongside usage data: tokens per dollar, seats per tier, utilization against plan limits. Renew on real numbers, not vendor summaries.
Anthropic team plan caps tokens at half what individual accounts deliver. Team hits 85% of cap consistently. Renegotiated before the next cycle.
Department-level cost allocation with rolling 12-month history. Chargeback reports formatted for FP&A models without additional formatting work.
Rachel's analyst runs allocation on the last business day of the month. Department heads receive their share automatically. Board AI slide in three minutes.
Integrations
Click any column header to sort. Filter by integration path to see vendors with similar integration shape.
All vendor data validated 2026-04-29.
| Vendor↕ | Path↑ | Cost data↕ | Granularity↕ | Plan required↕ | Setup↕ |
|---|---|---|---|---|---|
| ▸Anthropic API | Path 1 | Direct | Per key, workspace, model | Any · Admin role | ≤15 min |
| What we pull Cost and usage by API key, workspace, account, and model via the Admin API's Cost Report and Usage Report endpoints. Both return USD directly. What we don't get The Cost Report endpoint groups by workspace and description rather than by API key. Per-key dollar attribution is reconciled against Anthropic's published per-model pricing. | |||||
| ▸Claude.ai Enterprise | Path 1 | Direct | Per user, daily | Enterprise only | ≤15 min |
| What we pull Per-user spend (CSV, daily refresh) plus engagement metrics via the dedicated Analytics API. Activity logs available through the Compliance API. What we don't get This is a separate integration from the developer Anthropic API. It uses different keys. Team plan customers get CSV-only access (see separate Path 4 row). | |||||
| ▸Cursor | Path 1 | Direct | Per seat, daily | Enterprise only | ≤15 min |
| What we pull Per-seat spend (spendCents) and daily usage events from the Admin API, including filtered usage events for activity tracking and renewal decisions. What we don't get Seat subscription totals come from totalMembers × seat price or your invoices, not directly from the API. Teams plan ($40/user/month) gets a dashboard but no API access. | |||||
| ▸GitHub Copilot | Path 1 | Estimated seats × list price | Per seat | Business or Enterprise | ≤15 min |
| What we pull Per-seat activity from the new Copilot Metrics API (GA 2026-02-27). Seat count, activity frequency, and last active date. What we don't get GitHub does not return a dollar amount per seat. Cost is calculated as seat count × plan list price ($19 Business, $39 Enterprise). Customers with negotiated rates can override during setup. Seat/billing endpoints are still in public preview. | |||||
| ▸OpenAI | Path 1 | Direct | Per user, key, project, model | Any paid · Owner role | ≤15 min |
| What we pull Token-level usage and cost by model, project, API key, and user, refreshed daily via the Admin API. Per-user dollar attribution is computed from token counts × OpenAI's published per-model pricing. What we don't get OpenAI's Costs endpoint groups only by project and line item, so per-key and per-user dollar amounts are derived rather than direct from the vendor. | |||||
| ▸xAI / Grok | Path 1 | Direct | Per key, model | Any paid team | ≤15 min |
| What we pull USD billing time-series, invoices, balance, and configured spending limits via the Management API. Per-key segmentation visible in the Usage Explorer console. What we don't get Per-API-key segmentation dimensions are not exhaustively enumerated in the public reference. Some console-visible dimensions may not be in the API. | |||||
| ▸Alibaba Qwen | Path 2 | Direct | Per workspace, model | Any PAYG account | IT-coordinated |
| What we pull Cost via Alibaba Cloud BSS OpenAPI with daily granularity. InstanceID dimension can be parsed for per-workspace and per-model attribution. What we don't get Mainland (CNY) and International (USD) Alibaba Cloud accounts are billed separately and must be connected separately. Per-API-key cost requires parsing InstanceID strings and cross-referencing with the customer's API Key Management page. | |||||
| ▸AWS Bedrock | Path 2 | Direct | Per model, account | Any AWS account | IT-coordinated |
| What we pull Cost by model, service, and account via AWS Cost Explorer API or CUR-to-S3 with Athena queries. Per-model breakouts derived from USAGE_TYPE string parsing. What we don't get AWS does not expose MODEL as a first-class cost dimension. Per-model attribution depends on parsing USAGE_TYPE strings, which can change when new Bedrock models or usage-type names ship. | |||||
| ▸Azure OpenAI | Path 2 | Direct | Per deployment, subscription | EA / MCA preferred | IT-coordinated |
| What we pull Cost by subscription and deployment via Azure Cost Management. Per-deployment costs require consistent deployment tagging by your team. What we don't get The full Cost Details Report API requires Enterprise Agreement or Microsoft Customer Agreement. Pay-as-you-go customers fall back to the legacy usageDetails endpoint, which Microsoft is deprecating. Tokens are not on the cost path. | |||||
| ▸Google Gemini | Path 2 | Direct | Per project, model | Any GCP account | IT-coordinated |
| What we pull Cost by model, project, and billing period via BigQuery billing export. Per-model SKUs visible through Cloud Billing. What we don't get GCP attributes spend to the calling project, not to API keys. Per-API-key tracking requires running one project per team or per key. Setup requires enabling BigQuery billing export and granting read access. | |||||
| ▸Microsoft 365 Copilot | Path 2 | Estimated licenses × your rate | Per user activity | M365 Copilot license | IT-coordinated |
| What we pull Per-user activity dates per Copilot app via Microsoft Graph reports endpoint. License count and assignment data via Entra. What we don't get Microsoft does not return a dollar amount per seat. Cost is calculated as license count × your negotiated $/seat (entered during setup). User names are anonymized by default. A Global Administrator must enable real UPNs for reports. | |||||
| ▸Harvey | Path 3 | Calculated your contract rate | Per user, audit logs | Enterprise | Sales-led |
| What we pull Per-user usage history, query history, audit logs, and client-matter metadata via Harvey's developer API. What we don't get Harvey's API does not expose any dollar or cost field. Cost is calculated as observed seat count × the per-lawyer rate from your Order Form (entered during setup). API access must be provisioned through your Customer Success Manager. | |||||
| ▸Jasper | Path 3 | Manual from invoice | Per seat, API generations | Business · 12-mo min | ≤15 min |
| What we pull API generation counts via /v1/usage endpoint. Seat data via SCIM 2.0. Admin Usage Page CSV export covers 90 days. What we don't get Jasper has no billing API. Cost data must come from invoice upload. The Usage API only counts API requests, not in-product UI activity, so total seat utilization is partial. Pro plan customers cannot connect at all. | |||||
| ▸Mistral AI | Path 3 | Direct | Per workspace | Any · Admin role | ≤15 min |
| What we pull Cost data at workspace level via Mistral's Admin API. Workspace-level USD reporting with date-range filtering. What we don't get Mistral aggregates all keys in a workspace and bills at that level. There is no per-API-key dimension. We recommend customers adopt a one-key-per-workspace convention if per-key attribution matters. The Admin API page mentions billing/usage capability but publishes no concrete endpoint specs. | |||||
| ▸Perplexity for Teams | Path 3 | Manual from invoice | Per user via audit log | 50+ seats or Ent. Max | ≤15 min |
| What we pull Real-time per-user query events via the audit-log webhook. SCIM for seat provisioning data. What we don't get Perplexity has no REST API for billing, seats, or usage analytics. Cost data must come from invoice upload. Audit Logs are gated to Enterprise tenants with 50+ seats or at least one Enterprise Max user. Smaller customers can only export from the admin portal manually. | |||||
| ▸Claude.ai Team | Path 4 | Direct CSV | Per user, monthly | Team plan | Manual upload |
| What we pull Per-user usage data via CSV export from the Anthropic admin console. What we don't get Team plan customers have no programmatic API access. There is no Analytics API at this tier. Customers wanting API-level integration must upgrade to Enterprise (see separate Path 1 row). | |||||
| ▸Cohere | Path 4 | Manual from invoice | Per response tokens | Any plan | Manual upload |
| What we pull Per-response token counts via meta.billed_units on each chat or embeddings response. Aggregated invoice and usage data via CSV download from the Cohere dashboard. What we don't get Cohere has no documented billing or usage endpoint. The only programmatic signal is per-response token metadata. All aggregated cost data is dashboard-only and requires monthly CSV upload. | |||||
| ▸DeepSeek | Path 4 | Direct CSV ⓘ Compliance note | Per key, model | Any plan | Manual upload |
| What we pull Per-key and per-model cost data via monthly CSV export from platform.deepseek.com/usage. What we don't get DeepSeek's only programmatic surface is a single account-balance endpoint. All detail data must be exported manually. DeepSeek V4 launched 2026-04-24 (V4 Flash + V4 Pro SKUs). Legacy model IDs are flagged for deprecation. | |||||
| ▸Fireworks AI | Path 4 | Estimated tokens × rates | Per user (email) | Any paid | Manual upload |
| What we pull Per-user (by email) cost data via firectl CLI export. Token counts and accelerator-seconds available in CSV. What we don't get Fireworks does not support per-API-key attribution. If you need to allocate spend to a specific service account or workload-key, that data is not available. Cost in USD is computed from raw token counts × Fireworks' published rate card. Discounts (cached input, batch inference) must be applied client-side. | |||||
Showing 19 of 19 vendors
Granularity glossary. Per user = attributed to a named person (e.g., user.email). Per seat = attributed to a license slot. Per key, workspace, project, deployment, account = vendor-specific cost dimensions.
We re-check these quarterly. If a vendor ships an admin/billing API or removes a compliance blocker, we add them.